Clinical characteristics, treatment outcomes, and risk factors for mortality in post-neurosurgical central nervous system infections caused by Acinetobacter baumannii: a retrospective cohort study

Abstract

Background

Post-neurosurgical central nervous system infections (PCNSIs) caused by Acinetobacter baumannii pose a significant therapeutic challenge; however, comprehensive studies focusing on this specific scenario are limited. This study aims to analyze the clinical characteristics, treatment outcomes, and prognostic factors of patients with PCNSIs caused by Acinetobacter baumannii.

Methods

A retrospective cohort study was conducted at Guangdong Sanjiu Brain Hospital from January 2020 to June 2025. Patients diagnosed with Acinetobacter baumannii PCNSIs were included and stratified into carbapenem-resistant Acinetobacter baumannii (CRAB) and carbapenem-susceptible Acinetobacter baumannii (CSAB) groups. Clinical data, treatment regimens, and outcomes were analyzed. The primary outcome was 28-day all-cause mortality. Variables selected by LASSO regression were entered into univariate and multivariate Firth’s logistic regression models to identify risk factors associated with mortality.

Results

A total of 47 cases were included in the study, with 37 (78.7%) attributed to CRAB and 10 (21.3%) to CSAB. The 28-day mortality rate for CRAB infections was 37.8% (14/37), while no deaths occurred in the CSAB group (P = 0.022). In the CRAB group, Kaplan-Meier analysis indicated that non-polymyxin B-based regimens were associated with significantly higher mortality (70.5%) compared to polymyxin B-based regimens (10.0%; P < 0.001). Univariate analysis revealed that drainage performed following infection (OR = 0.087, 95% CI:0.008–0.502, P = 0.005), polymyxin B-based regimen (OR = 0.059, 95% CI:0.009–0.274, P < 0.001) and mechanical ventilation (OR = 4.399, 95% CI:1.127–19.002, P = 0.033) differed significantly between the groups. Multivariate analysis identified polymyxin B-based regimen as an independent factor associated with lower 28-day mortality (OR = 0.075, 95% CI: 0.008–0.402, P = 0.002).

Conclusion

CRAB-related PCNSIs are associated with significantly higher mortality compared to CSAB infections. Polymyxin B-based therapy may be associated with improved survival outcomes, while mechanical ventilation may be associated with an increased risk of mortality, warranting heightened clinical vigilance. Although post-infection drainage showed a protective effect in univariate analysis, it did not retain statistical significance as an independent predictor in the multivariable model, suggesting only a potential benefit that requires further validation.

Trial registration

Not applicable.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12879-026-13792-8.

Introduction

Post-neurosurgical central nervous system infections (PCNSIs) represent a formidable and devastating complication in neurosurgical practice, associated with significantly elevated mortality, substantial morbidity, and poor long-term functional outcomes, thereby imposing a heavy burden on patients, families, and healthcare systems [1, 2]. These infections are frequently linked to invasive procedures or conditions inherent to neurosurgery and critical care, such as external ventricular drainage, open cranial trauma, prolonged hospitalization, and mechanical ventilation [3–5]. Reported incidence rates vary widely, ranging from 0.5% to 21.4%, reflecting differences in surgical procedures, patient populations, and diagnostic criteria [5]. The management of PCNSIs is particularly challenging due to the sanctuary site of infection, the presence of foreign materials (e.g., shunts, drains), and the impaired host defense often seen in this critically ill population.

Among the diverse pathogens implicated in PCNSIs, Acinetobacter baumannii has emerged as a pathogen of paramount concern with increasing clinical prevalence [6, 7]. This Gram-negative coccobacillus possesses a remarkable and alarming capacity to develop resistance to multiple classes of antimicrobial agents, including last-resort carbapenems. The rise of CRAB has become a critical global health threat. Data from the China Antimicrobial Surveillance Network (CHINET) underscores the severity of this issue in the region, revealing a CRAB detection rate as high as 64.7% in 2024 [8]. The emergence and spread of CRAB clones have drastically narrowed the therapeutic arsenal, transforming otherwise manageable infections into clinical emergencies with limited options, especially within the vulnerable neurosurgical patient cohort [6, 9].

The management of CRAB-induced PCNSIs presents unique and profound challenges that extend beyond mere antimicrobial resistance. The blood-brain barrier (BBB) significantly restricts the penetration of many systemically administered antibiotics, complicating or preventing the achievement of therapeutic concentrations in the cerebrospinal fluid (CSF), which is crucial for eradicating infections [10]. Consequently, treatment strategies are forced to rely on last-resort agents with potential neurotoxicity, such as polymyxins (e.g., polymyxin B, colistin). These are often administered via innovative and invasive routes, including intrathecal or intraventricular injection, in combination with systemic therapy, in an attempt to bypass the BBB [11, 12]. However, the optimal treatment regimen—encompassing the choice of antimicrobials, their combination, the route and dosage of administration (particularly for polymyxins), and the duration of therapy—remains poorly defined and is primarily guided by limited retrospective evidence and expert opinion. Furthermore, the clinical effectiveness, safety profile, and pharmacodynamics of such aggressive regimens in the specific context of post-neurosurgical infections are not well characterized.

While prior studies have contributed valuable insights into the general risk factors for nosocomial infections and the evolving resistance patterns of Acinetobacter baumannii [7, 13], there is a notable paucity of research focusing specifically on its role and impact in PCNSIs. Detailed comparative analyses between CRAB and CSAB infections in the post-neurosurgical setting are particularly scarce. A granular understanding of the distinctive clinical profiles, host risk factors, treatment responses, and mortality predictors for these two groups is urgently needed. Such knowledge is essential to guide early clinical decision-making, inform antimicrobial stewardship, and ultimately improve patient prognosis.

Therefore, this retrospective cohort study was conducted to investigate and compare the clinical characteristics, anti-infective treatment strategies employed, and clinical outcomes of patients with Acinetobacter baumannii PCNSIs, with a specific focus on delineating the differences between CRAB and CSAB infections. We further aimed to identify independent risk factors associated with 28-day mortality. The ultimate goal is to provide contemporary clinical evidence that can inform more effective, timely, and targeted management strategies for this severe and life-threatening infection.

Methods

Study design and ethical considerations

This cohort study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (Supplementary Material 1) [14]. In accordance with national regulations “The Ethical Review of Biomedical Research Involving Humans (2016)” from the Chinese National Health and Family Planning Commission [15], the requirement for informed consent was waived as this study utilized retrospective, anonymized patient data. The Ethics Committee of Guangdong Sanjiu Brain Hospital granted the waiver.

Patients

We conducted a retrospective cohort study to investigate Acinetobacter baumannii PCNSIs at Guangdong Sanjiu Brain Hospital from January 2020 to June 2025. A total of 47 patients were enrolled based on strict inclusion and exclusion criteria. We compared the clinical features of two groups: 37 cases of CRAB PCNSIs and 10 cases of CSAB PCNSIs. Anti-infection regimens for CRAB PCNSIs were documented. Identification of Acinetobacter baumannii and antimicrobial susceptibility testing were performed on isolates from cerebrospinal fluid (CSF) specimens. The research process is illustrated in Fig. 1.

Fig. 1.

Study flow diagram. A total of 217 patients with positive CSF culture for Acinetobacter baumannii were initially screened. After exclusion, 47 patients with PCNSIs were included and stratified into CRAB (n = 37) and CSAB (n = 10) groups. The CRAB cohort was the primary focus of prognostic analysis

The inclusion criteria were as follows: (1) positive CSF bacterial culture identifying Acinetobacter baumannii; (2) confirmed diagnosis of PCNSI, meeting the diagnostic criteria established by expert consensus on PCNSIs [16]; (3) age ≥ 18 years; (4) complete relevant clinical data. The exclusion criteria were: (1) duplicate isolates from the same infection episode (only the first isolate from each infectious episode was included for analysis); (2) age < 18 years; (3) other bacteria were isolated from the cerebrospinal fluid culture; (4) patients not meeting the diagnostic criteria for intracranial infection.

Data collection

Medical records of all included patients were reviewed. Data extracted included demographic information, chronic diseases, reasons for craniotomy, prior use of antibiotics, intensive care unit (ICU) admission, length of stay in the ICU, mechanical ventilation, immunocompromised status, CSF leakage, number of operations, use of artificial material implantation, drainage indwelling, co-infection status, postoperative time to PCNSIs (days), and clinical outcome. CSF examination results were recorded when cultures were positive. The primary outcome was 28-day all-cause mortality. Secondary outcomes included clinical cure, clinical remission, and clinical failure at 28 days, as defined in the Definitions section (see below).

Regarding the management of neurosurgical hardware, institutional practice includes routine removal of external ventricular drains (EVDs) within one week of insertion in the absence of infection. For lumbar drains, the indwelling period may be extended to 2–3 weeks. In patients with postoperative intracranial infection, modified EVDs are often placed and may be retained for one month or longer to facilitate infection control and intracranial pressure monitoring. However, due to the retrospective nature of this study, precise data on the timing and systematic performance of hardware removal were not available for all patients and were therefore not included in the quantitative analysis.

Antibiotic susceptibility testing

The minimum inhibitory concentrations (MICs) for acinetobacter baumannii isolates of 14 antibiotics including ceftazidime, ciprofloxacin, cefoperazone/sulbactam, doxycycline, cefepime, imipenem, levofloxacin, meropenem, minocycline, polymyxin B, trimethoprim/sulfamethoxazole, tigecycline, ticarcillin/clavulanic acid, tobramycin, piperacillin/tazobactam, were tested using the broth microdilution method according to CLSI guidelines [17]. The EUCAST breakpoint was used for polymyxin B and tigecycline [18]. The control strain used was Escherichia coli ATCC 25,922. Furthermore, novel anti-CRAB agents such as cefiderocol and sulbactam-durlobactam were not included in routine susceptibility testing, as they were not clinically available at our institution during the study period.

Definition

In this study, the clinical outcomes were categorized as follows (consistent with the Data Collection section): clinical cure (the resolution of symptoms/signs and antibiotics-free); clinical remission (the partial resolution of symptoms/signs but not antibiotics-free); and clinical failure (the persistence or progression of symptoms/signs, and/or the development of a new symptoms/signs suggestive of new infection) [19]. The primary outcome was 28-day all-cause mortality, defined as death from any cause within 28 days following the first positive CSF culture. The patients were defined as immunosuppressed if they had HIV or AIDS, were transplant recipients, had received chemotherapy within the previous 6 weeks, had received systemic steroids (equivalent to or greater than 20 mg of prednisone) for 2 weeks, or had been treated with other immunosuppressive agents within 2 weeks before hospitalization [20].

A polymyxin B-based regimen was defined as any antibiotic combination that included polymyxin B administered via intravenous (IV) and/or intrathecal/intraventricular (IT/IVT) route, regardless of the presence of additional systemic agents. At our institution, polymyxin B is administered according to the standardized regimen aligned with the current package insert: intravenous dose 500,000 U q12h; intraventricular/intrathecal dose 50,000 U qd. Only patients who received this standard dosage for ≥ 48 h were included in the “polymyxin B-based regimen” group. Empirical therapy was defined as antibiotics administered prior to the availability of CSF culture and susceptibility results. Targeted therapy was defined as antibiotic treatment adjusted based on culture and susceptibility results. Timely targeted therapy was defined as initiation of an active agent against CRAB within 48 h of the positive culture result.

According to the chinese expert consensus on diagnosis and treatment of neurosurgical central nervous system infections (2021) [16], the diagnostic criteria for PCNSIs are: (1) Clinical manifestations, including systemic inflammatory response (e.g., fever > 38 °C or hypothermia < 36 °C), altered mental status, signs of increased intracranial pressure, and meningeal irritation; (2) Laboratory findings showing leukocytosis (> 10.0 × 10⁹/L) with neutrophilia; (3) Cerebrospinal fluid (CSF) analysis revealing an opening pressure > 200 mmH₂O, turbid or purulent appearance, elevated white cell count (> 100 × 10⁶/L) with neutrophil predominance (> 70%), decreased glucose concentration (< 2.2 mmol/L), and a CSF-to-serum glucose ratio < 0.4; (4) Supportive neuroimaging features such as meningeal enhancement or abscess formation. A definite etiological diagnosis requires meeting all clinical diagnostic criteria plus positive microbiological culture or smear from CSF, wound secretion, drain tip, or implanted device, excluding contamination or colonization.

Based on these diagnostic criteria, infection was distinguished from colonization based on the presence of clinical signs and symptoms of meningitis/ventriculitis (fever, altered mental status, meningeal irritation) combined with CSF pleocytosis, elevated protein, and decreased glucose, in addition to a positive CSF culture. Cases with a positive CSF culture but without corresponding clinical or CSF findings suggestive of active infection were considered colonization and were excluded from this study.

Statistical analysis

Data were analyzed using IBM^® SPSS software (version 22.0). Descriptive statistics for continuous and categorical variables were summarized as means (standard deviations, SDs) or medians (25th and 75th percentiles) and numbers (percentages), respectively. The student’s t-test was used to compare normally distributed data, and the nonparametric Mann-Whitney U test was used to compare non-normally distributed data. Fisher’s exact test and χ2 test were used for statistical evaluation, and along with Kaplan-Meier survival analysis. The survival rates of different groups were compared by log-rank test. Logistic regression analysis showed factors associated with patient survival (P < 0.05), and odds ratios (OR) with 95% confidence interval (CI). P < 0.05 was considered statistically significant. Complete-case analysis was performed because missing data were < 2% for all variables.

To address the challenges posed by high-dimensional data and multicollinearity among variables, we employed least absolute shrinkage and selection operator (LASSO) regression. A total of 23 clinical and laboratory variables were considered, including: Age, Sex (male), Diabetes, Hypertension, Hypoproteinemia, Tumor, Days in hospital, ICU admission, Mechanical ventilation, Immunocompromised status, Number of operations (≥ 2), CSF leakage, Post-infectious imaging revealing intraventricular empyema, Initial empiric therapy covering Acinetobacter baumannii, Polymyxin B-based regimen, Antibiotic regimen with ≥ 3 agents, Initiating targeted therapy ≤ 48 h post-culture, Drainage performed following infection, CSF white blood cells, CSF glucose, CSF chloride, CSF protein, and CSF lactate. LASSO regression analysis was performed with R version 4.3.1 (https://www.R-project.org) and EasyR (https://www.easyr.cc Solutions, Inc., Shanghai). Ten-fold cross-validation was applied to select the optimal penalty parameter (λ) that minimized the binomial deviance. Variables with non-zero coefficients at the optimal λ were retained for further logistic regression analysis.

Given the relatively low number of outcome events (28-day mortality, n = 14) and the potential for small-sample bias in conventional logistic regression, we performed a bias-reduced logistic regression using the Firth penalized likelihood method as a sensitivity analysis. This approach helps to mitigate the first-order bias of maximum likelihood estimates and provides more reliable inferences in datasets with rare events or separation. The analysis was conducted using the R software (version 4.5.2). Variables selected by LASSO regression were entered into the Firth model to confirm the robustness of the identified independent predictors. Odds ratios (OR) and 95% confidence intervals (CI) were calculated based on the profile penalized likelihood.

To assess the robustness of our primary findings, we conducted two sensitivity analyses. First, we re-ran the Firth’s penalized logistic regression using the set of variables selected by the less stringent lambda.min criterion (8 variables: mechanical ventilation, drainage following infection, CSF white blood cells, age, sex, polymyxin B-based regimen, initiating targeted therapy ≤ 48 h, and days in hospital). Second, we performed a restricted analysis excluding patients with incomplete source control (defined as lack of post-infection drainage, n = 7) to evaluate whether the association of polymyxin B-based regimen with mortality was consistent.

Results

Clinical characteristics

A total of 47 patients with Acinetobacter baumannii PCNSIs were ultimately included in this study. Our study investigated the clinical characteristics of Acinetobacter baumannii PCNSIs, and compared the clinical features between CRAB and CSAB (Table 1). All the 47 patients had undergone neurosurgical interventions, including craniotomy for vascular diseases (22/47, 46.8%), tumors (12/47, 25.5%), trauma (11/47, 23.4%), and hydrocephalus (2/47, 4.2%). A considerable proportion of the patients had comorbidities, with hypoproteinemia being the most prevalent (30/47, 63.8%). In cases of co-infection, pulmonary infection accounted for the largest proportion (28, 59.6%). The average time from the last surgery to positive CSF culture was 16 days. A total of 83.0% (39/47) patients had a history of indwelling catheters after their most recent neurosurgical surgery. The 28-day mortality rate was 29.8% (14/47) (Table 2). The cerebrospinal fluid examination revealed characteristics of markedly elevated white blood cell counts and significantly reduced glucose concentration. Clinical outcomes of the entire cohort, including 28-day mortality and treatment response categories (clinical cure, remission, and failure), are summarized in Table 2.

Table 1.

The clinical features comparison between CRAB and CSAB PCNSI cases

	Total (n = 47)	CRAB (n = 37)	CSAB (n = 10)	P-value
Age (y)	50 (40, 56)	52 (40.5, 56.0)	40.5 (33.8, 67.3)	0.677
Sex (male)	29 (61.7)	23 (62.2)	6 (60.0)	> 0.999
Diabetes	7 (14.9)	4 (10.8)	3 (30.0)	0.155
Hypertension	17 (36.2)	15 (40.5)	2 (20.0)	0.289
Hypoproteinemia	30 (63.8)	26 (70.3)	4 (40.0)	0.136
Days in hospital (d)	43 (27, 74)	38 (25, 61)	70.5 (34.5, 82.5)	0.077
ICU admission	24 (51.1)	19 (51.4)	5 (50.0)	> 0.999
Length of stay in the ICU (d)	8 (5.3, 10.8)	9 (5, 12)	6 (4.5, 9.5)	0.475
Mechanical ventilation	13 (27.7)	13 (36.1)	0	0.043
Immunocompromised status	12 (25.5)	9 (24.3)	3 (30.0)	0.700
Reasons for the operation
Trauma	11 (23.4)	9 (24.3)	2 (20.0)	> 0.999
Tumor	12 (25.5)	8 (21.6)	4 (40.0)	0.251
Vascular diseases	22 (46.8)	20 (54.0)	2 (20.0)	0.079
Hydrocephalus	2 (4.2)	0	2 (20.0)	0.042
Number of operations (≥ 2)	19 (40.4)	13 (35.1)	6 (60.0)	0.276
CSF leakage	5 (10.6)	2 (5.4)	3 (30.0)	0.057
Postoperative time to PCNSIs (d)	16 (11, 25)	20 (12.5, 26.5)	13 (8, 21.5)	0.057
Post-infectious imaging reveals an intraventricular empyema	24 (51.1)	19 (51.4)	5 (50.0)	> 0.999
Initial empiric therapy covers Acinetobacter baumannii	10 (21.3)	4 (10.8)	6 (60.0)	0.003
Initiating targeted therapy ≤ 48 h post-culture	34 (72.3)	24 (64.9)	10 (100.0)	0.043
drainage performed following infection	38 (81.9)	30 (81.1)	8 (80.0)	> 0.999
Implantation of artificial material before PCNSI
Ommaya pump	1 (2.1)	0	1 (10.0)	> 0.999
Artificial meninges	3 (6.3)	2 (5.4)	1 (10.0)	0.521
Ventriculoperitoneal shunt	3 (6.3)	2 (5.4)	1 (10.0)	0.521
Indwelling drainage before PCNSI
External ventricular drain	30 (63.8)	23 (62.1)	7 (70.0)	0.727
Subcutaneous Drainage	4 (8.5)	3 (8.1)	1 (10.0)	> 0.999
Drainage of ventricular and subdural fluid	3 (6.3)	1 (2.7)	2 (20.0)	0.110
Lumbar CSF drainage	2 (4.2)	2 (5.4)	0	> 0.999
Co-infections
Bloodstream infection	2 (4.2)	2 (5.4)	0	> 0.999
Pulmonary infection	28 (59.5)	22 (59.4)	6 (60.0)	> 0.999
Pulmonary infection + Urinary tract infection	4 (8.5)	3 (8.1)	1 (10.0)	> 0.999
Bloodstream infection + Pulmonary infection	9 (19.1)	8 (21.6)	1 (10.0)	0.660
Bloodstream infection + Pulmonary infection + Urinary tract infection	1 (2.1)	1 (2.7)	0	> 0.999
CSF tests
CSF white blood cells (/µL)	8840 (2640, 24000)	8840 (3520, 22700)	13,320 (1400, 32600)	0.948
CSF glucose (mmol/L)	0.1 (0.1, 0.7)	0.1 (0.1, 0.6)	0.4 (0.075, 1.625)	0.498
CSF chloride (mmol/L)	115.7 (111.1, 123.4)	115.4 (109.95, 120.5)	122.8 (112.1, 128.8)	0.069
CSF protein (g/L)	4.58 (3.46, 8.39)	5.21 (3.77, 10.75)	3.62 (1.67, 4.88)	0.017
CSF lactate (mmol/L)	12.95 (9.7, 15.3)	13.02 (10.0, 15.7)	10.9 (5.5, 14.4)	0.125

Table 2.

Clinical outcomes of CRAB and CSAB PCNSI cases

	Total (n = 47)	CRAB (n = 37)	CSAB (n = 10)	P-value
28-day mortality	14 (29.8)	14 (37.8)	0	0.022
Clinical cure (%)	7 (14.9)	4 (10.8)	3 (30.0)	0.155
Clinical remission (%)	22 (46.8)	15 (40.5)	7 (70.0)	0.154
Clinical failure (%)	18 (38.3)	18 (48.6)	0	0.006

The study performed comparative analysis between the CRAB group and the CSAB group, as depicted in Table 1. The findings demonstrated that patients within the CRAB cohort had a significantly higher 28-day mortality (37.8% vs. 0%, P = 0.022), a higher proportion requiring mechanical ventilation (36.1% vs. 0%, P = 0.043), a lower rate of initial treatment coverage for pathogenic bacteria (10.8% vs. 60.0%, P = 0.003), and a lower proportion of timely targeted treatment within three days (64.9% vs. 100.0%, P = 0.043). In cerebrospinal fluid-related tests, the protein level in the CRAB group was significantly higher than that in the CSAB group (P = 0.017).

The drug susceptibility results are shown in Table 3. Among them, cerebrospinal fluid cultures from 37 (78.7%) patients showed infections with carbapenem-resistant Acinetobacter baumannii (CRAB), while 10 (21.3%) patients had infections with carbapenem-sensitive Acinetobacter baumannii (CSAB).

Table 3.

Acinetobacter baumannii antibiotic sensitivity rates

Organism	CAZ	CIP	CSL	DOX	FEP	IPM	LEV	MEM	MNO	POLY	SXT	TGC	TIC/CLA	TOB	TZP
AB	8	6	9	10	9	10	7	10	17	42	19	25	8	10	8

CAZ ceftazidime, CIP ciprofloxacin, CSL cefoperazone/Sulbactam, DOX doxycycline, FEP cefepime, IPM imipenem, LEV levofloxacin, MEM meropenem, MNO minocycline, POLY polymyxin B, SXT trimethoprim/sulfamethoxazole, TGC tigecycline, TIC/CLA ticarcillin/clavulanic acid, TOB tobramycin, TZP piperacillin/tazobactam, AB Acinetobacter baumannii

Strategies and outcomes of CRAB PCNSIs

Table 4 summarizes the anti-infective regimens and corresponding clinical outcomes of patients infected with CRAB. A majority of patients (86.5%, 32/37) received combination therapy comprising two or more antimicrobial agents. Among them, 40.5% showed clinical remission, while 48.7% experienced clinical failure. Of the CRAB-infected patients, 54.1% were treated with polymyxin B. Within this subgroup, 90% received combined intravenous and intrathecal/intravenous administration of polymyxin B, one patient received intrathecal injection only, and one patient received intravenous administration only (clinical cure). Among the 37 CRAB patients, those who received polymyxin B-based regimens (n = 20) had a significantly lower clinical failure rate (20.0%, 4/20) than those treated with non-polymyxin B-based regimens (82.4%, 14/17; P = 0.003). Within the non-polymyxin B-based group, sulbactam-based regimens showed a failure rate of 62.5% (5/8), while all patients receiving other combination regimens (n = 9) experienced clinical failure. The Kaplan-Meier survival analysis confirmed a significant survival advantage for the polymyxin B-based group (Fig. 2, log-rank P < 0.001).

Table 4.

Summary of anti-infection treatment and outcome of 37 CRAB PCNSIs cases

Anti-infection therapy	Number (%)	28-day
		clinical cure (%)	clinical remission (%)	clinical failure (%)
Polymyxin B-based regimens	20 (54.1)	4 (20.0)	12 (60.0)	4 (20.0)
Non-polymyxin B-based combination regimen	17 (45.9)	0	3 (17.6)	14 (82.4)
Sulbactam-based regimens	8 (21.6)	0	3 (37.5)	5 (62.5)
Other regimens	9 (24.3)	0	0	9 (100.0)
Total	37 (100.0)	4 (10.8)	15 (40.5)	18 (48.7)

*Comparison of clinical failure rates between the polymyxin B-based and non-polymyxin B-based regimens yielded a P = 0.003 (Fisher’s exact test)

Fig. 2.

Kaplan-Meier survival curves of the patients treated with Polymyxin B-based combination regimen vs. Non-polymyxin B-based combination regimen (log-rank test, P < 0.001)

To assess potential confounding by indication, we compared the baseline clinical characteristics between patients who received polymyxin B-based regimens (n = 20) and those who received non-polymyxin B-based regimens (n = 17) within the CRAB cohort (Table 5). The two groups were generally well-balanced in terms of demographics, comorbidities, disease severity indicators (ICU admission, mechanical ventilation), and baseline CSF parameters. However, patients in the polymyxin B-based group had a significantly longer median hospital stay (45.5 vs. 27 days, P = 0.014) and were more likely to have undergone drainage procedures following infection (95.0% vs. 64.7%, P = 0.033) compared to the non-polymyxin B group. Notably, the polymyxin B-based group also had a significantly higher proportion of patients who received initiating targeted therapy ≤ 48 h post-culture (90.0% vs. 35.3%, P = 0.001), reflecting a more timely escalation to effective therapy. Despite these differences, the 28-day mortality was strikingly lower in the polymyxin B-based group (10.0% vs. 70.5%, P < 0.001), suggesting that the survival benefit associated with polymyxin B-based therapy is robust and unlikely to be explained by more favorable baseline characteristics.

Table 5.

Baseline characteristics of CRAB patients stratified by use of polymyxin B-based therapy

	Total (n = 37)	Polymyxin B-based combination regimen (n = 20)	Non-polymyxin B-based combination regimen (n = 17)	P-value
Age (y)	52 (40.5, 56.0)	50 (40, 56)	54 (43, 61)	0.572
Sex (male)	23 (62.2)	11 (55.0)	12 (70.6)	0.498
Diabetes	4 (10.8)	3 (15.0)	1 (5.9)	0.609
Hypertension	15 (40.5)	9 (45.0)	6 (35.3)	0.738
Hypoproteinemia	26 (70.3)	14 (70.0)	12 (70.6)	> 0.999
Tumor	8 (21.6)	2 (10.0)	6 (35.3)	0.109
Days in hospital (d)	38 (25, 61)	45.5 (33.3, 80.0)	27 (9.5, 43.0)	0.014
ICU admission	19 (51.5)	10 (50.0)	9 (52.9)	> 0.999
Length of stay in the ICU (d)	1.5 (0, 9.25)	5 (0, 10)	1 (0, 7)	0.870
Mechanical ventilation	13 (35.1)	6 (30.0)	7 (41.2)	0.512
Immunocompromised status	9 (24.3)	5 (25.0)	4 (23.5)	> 0.999
Number of operations (≥ 2)	13 (35.1)	8 (40.0)	5 (29.4)	0.731
CSF leakage	2 (5.4)	0	2 (11.8)	0.204
Post-infectious imaging reveals an intraventricular empyema	19 (51.4)	12 (60.0)	7 (41.2)	0.330
Drainage performed following infection	30 (81.1)	19 (95.0)	11 (64.7)	0.033
Initial empiric therapy covers Acinetobacter baumannii	4 (10.8)	3 (15.0)	1 (5.6)	0.609
Initiating targeted therapy ≤ 48 h post-culture	24 (64.9)	18 (90.0)	6 (35.3)	0.001
CSF tests
CSF white blood cells (/µL)	8840 (3520, 22700	10,410 (4235, 28800)	8840 (2920, 14004)	0.442
CSF glucose (mmol/L)	0.1 (0.1, 0.6)	0.1 (0.1, 0.5)	0.1 (0.1, 0.9)	0.845
CSF chloride (mmol/L)	115.4 (109.9, 120.5)	116.35 (111.5, 120.6)	115.4 (108, 121.4)	0.357
CSF protein (g/L)	5.2 (3.8, 10.7)	5.9 (3.1, 10.7)	5.2 (3.8, 11.7)	0.869
CSF lactate (mmol/L)	13.0 (10.0, 15.7)	12.5 (10.5, 15.3)	13.7 (9.0, 18.0)	0.988
28-day mortality	14 (37.8)	2 (10.0)	12 (70.6)	< 0.001

Among the 37 CRAB patients, only 4 (10.8%) received empirical therapy covering CRAB; all four initially received sulbactam-based regimens. After culture results, three were transitioned to polymyxin B-based combination therapy, while one continued on a sulbactam-based regimen. Overall, 24 patients (64.9%) received targeted therapy within 48 h, of whom 18 (75.0%) received a polymyxin B-based regimen. In contrast, among the 13 patients without timely targeted therapy (> 48 h), only 2 (15.4%) received polymyxin B-based therapy, and 11 (84.6%) died.

LASSO regression analysis

Among the 23 variables entered into the LASSO regression model, ten-fold cross-validation identified the optimal λ values of 0.05 (lambda.min) and 0.12 (lambda.1se), corresponding to the model with minimum cross-validated error and the most parsimonious model, respectively (Fig. 3). At λ = 0.05, the following variables were selected: mechanical ventilation, drainage performed following infection, CSF white blood cells, Age, Sex, Polymyxin B-based regimen, Initiating targeted therapy ≤ 48 h post-culture, and Days in hospital. At the more stringent λ = 0.12, which retained four variables: mechanical ventilation, drainage performed following infection, polymyxin B-based regimen, and age. The coefficient shrinkage paths of all predictors across varying λ values are displayed in Fig. 4. Consequently, we included the 4 variables identified by lambda.1se in subsequent univariate and multivariate Firth’s logistic regression models.

Fig. 3.

Ten-fold cross-validation plot for selecting the optimal λ value in the LASSO regression model based on minimum mean cross-validated error

Fig. 4.

LASSO path plot illustrating coefficient shrinkage of predictors as the penalty parameter (λ) increases

Risk factors of the Acinetobacter baumannii PCNSIs 28-day mortality

The overall 28-day all-cause mortality rate in the study cohort was 29.8% (14/47). Table 6 presents the Firth’s logistic regression analysis of mortality risk factors. Univariate analysis revealed that drainage performed following infection (OR = 0.087, 95% CI:0.008–0.502, P = 0.005), polymyxin B-based regimen (OR = 0.059, 95% CI:0.009–0.274, P < 0.001) were protective against 28-day mortality, whereas mechanical ventilation (OR = 4.399, 95% CI:1.127–19.002, P = 0.033) was associated with increased mortality. Multivariable logistic regression analysis identified polymyxin B-based regimen as an independent predictor associated with lower 28-day mortality (OR: 0.075, 95% CI:0.008–0.402, P = 0.002).

Table 6.

Risk factors for 28-day mortality in CRAB PCNSI patients: Firth’s logistic regression analysis

	Non-survivor (n = 14)	Survivor (n = 23)	Univariate analysis			Multivariate analysis
			OR	95%CI	P-value	OR	95% CI	P-value
Age	55 (50.5, 64.25)	43 (40, 55)	1.057	0.997–1.134	0.062
Mechanical Ventilation	8 (57.1)	5 (21.7)	4.399	1.127–19.002	0.033	5.769	0.912–64.603	0.063
Polymyxin B-based regimen	2 (14.3)	18 (78.3)	0.059	0.009–0.274	< 0.001	0.075	0.008–0.402	0.002
Drainage performed following infection	8 (57.1)	22 (95.7)	0.087	0.008–0.502	0.005	0.169	0.007–1.962	0.163

Sensitivity analyses

Sensitivity analyses confirmed the robustness of the primary findings (Table 7). Using the lambda.min-selected variables (8 variables), Firth’s penalized logistic regression continued to identify polymyxin B-based regimen as the only independent factor associated with 28-day mortality (OR = 0.135, 95% CI:0.010–0.914, P = 0.040). In the restricted analysis excluding patients without post-infection drainage (n = 7, leaving 30 patients with 10 deaths), polymyxin B-based regimen remained significantly associated with lower mortality (OR = 0.086, 95% CI: 0.007–0.584, P = 0.011). These results support the stability of our primary conclusions.

Table 7.

Sensitivity analyses for the association of polymyxin B-based regimen with 28-day mortality

Analysis	n (deaths)	OR (95% CI)	P-value
Primary analysis (lambda.1se, 4 variables)	37 (14)	0.075 (0.008–0.402)	0.002
Sensitivity analysis 1 (lambda.min, 8 variables)	37 (14)	0.135 (0.010–0.914)	0.040
Sensitivity analysis 2 (restricted, excluding no drainage, 3 variables)	30 (10)	0.086 (0.007–0.584)	0.011

Discussion

This retrospective study analyzed the clinical characteristics, treatment strategies, and prognostic factors in 47 patients with PCNSIs caused by Acinetobacter baumannii, with a specific focus on the CRAB cohort. The key findings were: (1) The 28-day mortality rate was significantly higher in the CRAB group compared to the CSAB group (37.8% vs. 0%, P = 0.022); (2) Among CRAB-infected patients, those treated with a polymyxin B-based combination regimen had significantly lower mortality than those on non-polymyxin B-based regimens (10.0% vs. 70.6%, P < 0.001), despite having longer hospital stays and higher rates of post-infection drainage; (3) In univariate analysis, both drainage performed following infection and polymyxin B-based regimen were significantly associated with reduced mortality, whereas mechanical ventilation was associated with increased mortality. Multivariable Firth’s logistic regression identified polymyxin B-based regimen as an independent protective factor against 28-day mortality (OR = 0.075, 95% CI: 0.008–0.402, P = 0.002). Collectively, these results indicate that polymyxin B-based therapy and timely post-infection drainage may be associated with improving outcomes, while mechanical ventilation may represent an important risk marker that identifies patients requiring heightened clinical vigilance.

The markedly higher mortality observed in the CRAB group aligns with the multidrug-resistant nature of this pathogen and the limited penetration of many systemic antibiotics across the blood-brain barrier [10, 20]. Acinetobacter baumannii, particularly CRAB strains, often exhibits resistance to multiple antimicrobial classes which includes carbapenems, drastically narrowing therapeutic options [11, 12, 21]. In our cohort, only polymyxin B demonstrated a high in vitro susceptibility rate (100%), whereas resistance to other common agents like carbapenems and β-lactam/β-lactamase inhibitor combinations was prevalent. This microbiological profile underscores the reliance on polymyxins as a cornerstone of therapy for CRAB CNS infections, a finding consistent with current IDSA and Chinese guidelines [11, 12].

A critical finding of this study is that polymyxin B-based regimens were associated with lower 28-day mortality in CRAB PCNSIs. The majority of patients in this group (90%) received combination intravenous and intrathecal/intraventricular administration, a strategy that bypasses the blood-brain barrier to achieve therapeutic concentrations directly at the infection site [22–25]. This likely explains the regimen’s success despite the inherent difficulties in treating CNS infections. Importantly, as shown in Table 5, patients receiving polymyxin B-based therapy had comparable baseline characteristics to those receiving non-polymyxin B regimens in terms of demographics, comorbidities, and disease severity indicators (ICU admission, mechanical ventilation). The longer hospital stay and higher rate of post-infection drainage in the polymyxin B group would typically be associated with worse, not better, outcomes, yet this group demonstrated dramatically lower mortality. Furthermore, patients in the polymyxin B-based group were significantly more likely to receive targeted therapy within 48 h of culture positivity, reflecting a more timely escalation to effective therapy. This disparity strengthens the evidence that the observed survival benefit is attributable to the treatment itself rather than selection bias, and underscores the critical importance of rapid diagnostic confirmation (rapid molecular diagnostics, such as multiplex PCR or mNGS) and prompt initiation of effective antimicrobial therapy. Notably, this timely escalation was achieved despite the low overall rate of effective empirical therapy (10.8%), highlighting the essential role of rapid microbiological diagnosis and prompt adjustment to appropriate agents.

However, we acknowledge a significant limitation: with 90% of polymyxin B-treated patients receiving combined systemic and local administration, our data cannot statistically disentangle the independent contribution of the intrathecal/intraventricular route. Whether local administration provides additional benefit beyond systemic therapy alone remains an unresolved question requiring prospective investigation. Nevertheless, our real-world experience supports current expert recommendations favoring combined systemic and local polymyxin therapy for severe CRAB ventriculomeningitis [11, 22–25].

Univariate analysis identified drainage performed following infection as a significant protective factor (OR = 0.087, 95% CI: 0.008–0.502, P = 0.005), although it did not retain statistical significance in the multivariable model (OR = 0.169, 95% CI: 0.007–1.962, P = 0.163). This attenuation may reflect collinearity with polymyxin B-based therapy, as patients receiving polymyxin B were significantly more likely to undergo drainage (95.0% vs. 64.7%, P = 0.033). From a mechanistic perspective, external ventricular drainage serves multiple therapeutic purposes in CNS infections: it allows for removal of purulent CSF, reduces bacterial burden, lowers intracranial pressure, and provides a route for intraventricular antibiotic administration [3, 4]. Our findings, while not establishing independence in multivariable analysis, support the clinical practice of combining effective antimicrobial therapy with prompt CSF drainage when indicated. Nevertheless, given the lack of statistical significance in the multivariable model, the potential benefit of post-infection drainage should be interpreted cautiously, and its role as an independent adjunctive intervention remains to be confirmed in future studies. The high rate of drainage in survivors (95.7%) compared to non-survivors (57.1%) underscores its potential importance as an adjunctive intervention. Additionally, although we describe institutional practices regarding hardware removal, the lack of systematically collected data on removal timing precludes any analysis of its impact on outcomes, which should be addressed in future prospective studies.

Mechanical ventilation was significantly associated with increased mortality in univariate analysis (OR = 4.399, 95% CI: 1.127–19.002, P = 0.033) and showed a trend toward significance in the multivariable model (OR = 5.769, 95% CI: 0.912–64.603, P = 0.063). The loss of statistical significance in the adjusted model may be due to the limited sample size and the relatively low number of events (n = 14), which reduces power to detect independent effects. Mechanically ventilated patients represent a subgroup with greater disease severity, prolonged ICU stays, and increased exposure to nosocomial pathogens. From a clinical perspective, the presence of mechanical ventilation should alert clinicians to the heightened risk of poor outcomes and may justify more aggressive diagnostic and therapeutic interventions, including earlier consideration of polymyxin-based therapy and drainage procedures.

Our findings are partly consistent with those of Tian et al. [7], who also reported high mortality and treatment challenges in post-neurosurgical intracranial infections caused by A. baumannii. However, our study extends the literature by providing a detailed comparison between CRAB and CSAB infections, systematically evaluating the impact of polymyxin B-based therapy while controlling for baseline differences (Table 5), and employing Firth’s penalized logistic regression to address small-sample bias. The convergence of our results with prior studies supporting intrathecal/intraventricular polymyxin administration [22–25] reinforces the evidence base for this approach in multidrug-resistant A. baumannii meningitis/ventriculitis.

The clinical implications of our study are threefold. First, in settings where novel anti-CRAB agents such as cefiderocol or sulbactam-durlobactam are unavailable—which remains the reality for many centers globally—optimizing the use of polymyxin B, particularly via combined systemic and local administration, represents a critical strategy for improving survival. Second, our data support the integration of prompt post-infection drainage into the management algorithm for CRAB PCNSIs, as it may enhance bacterial clearance and provide a conduit for local therapy. Third, mechanically ventilated patients should be recognized as a high-risk subgroup warranting enhanced monitoring and early aggressive intervention.

However, limitations must be acknowledged. First, the retrospective, single-center design inherently carries risks of selection bias and unmeasured confounding. The sample size, particularly of the CSAB group (n = 10), limits the statistical power for some comparisons and more extensive multivariate modeling. Second, the definitions of “clinical cure”, “remission” and “failure”, while based on literature, are subject to retrospective assessment bias. Third, we did not assess long-term neurological outcomes or relapse rates, focusing instead on 28-day mortality as a primary, unequivocal endpoint. Fourth, microbiological characterization was limited to phenotypic susceptibility testing; molecular data on carbapenemase genes (e.g., OXA-type) and strain typing were unavailable, precluding analysis of clonal outbreaks versus polyclonal endemicity. Fifth, we did not systematically evaluate prior colonization status, which may influence empirical therapy choices and outcomes; the high rate of pulmonary co-infection/colonization (59.4% in the CRAB group) suggests that multisite surveillance cultures could potentially guide earlier targeted therapy, but this hypothesis requires prospective validation. Sixth, cerebrospinal fluid drug concentrations were not measured, limiting our ability to correlate pharmacokinetic parameters with clinical response. Finally, being from a single specialized neurosurgical center, the findings may not be fully generalizable to all hospital settings.

Additionally, we acknowledge the potential for confounding by indication, a common challenge in observational studies of treatment effects. Patients receiving polymyxin B-based therapy may have differed from those receiving non-polymyxin B regimens in unmeasured ways that influenced both treatment selection and outcomes. For example, clinicians may have reserved polymyxin B for more severely ill patients, which would bias results toward the null; conversely, if polymyxin B was preferentially used in patients with fewer contraindications, this could bias results away from the null. Although we observed no significant differences in baseline disease severity indicators (ICU admission, mechanical ventilation) between the two treatment groups (Table 5), residual confounding from unmeasured factors—such as prior antibiotic exposure, source control adequacy, or institutional practice patterns—cannot be excluded. Our findings should therefore be interpreted as hypothesis-generating rather than definitive evidence of treatment superiority.

Another important consideration pertains to time-dependent variables. In our analysis, variables such as ‘drainage performed following infection’ and ‘initiating targeted therapy ≤ 48 h post-culture’ were treated as binary (yes/no) indicators rather than time-varying covariates. The exact timing of drainage initiation relative to infection onset, the duration of drainage, and the temporal relationship between therapy escalation and clinical response were not captured. This simplification may obscure important temporal dynamics—for instance, drainage performed very early after infection onset might have a different impact than drainage delayed by several days. Similarly, the 48-hour threshold for targeted therapy, while clinically practical, does not capture the precise hour of effective therapy initiation. Future studies with more granular time-to-event data should employ time-dependent Cox models to better characterize these relationships.

Regarding model stability, we acknowledge that our primary analysis was limited by a low number of outcome events (n = 14 deaths in the CRAB cohort). Although we employed Firth’s penalized likelihood regression—a method specifically designed to reduce first-order bias in small samples with rare events or separation—and used a parsimonious model with only four predictors selected by the conservative lambda.1se criterion, the limited event count still constrains the precision of our estimates (reflected in wide confidence intervals, e.g., OR for polymyxin B: 95% CI 0.008–0.402). While the sensitivity analyses using alternative variable selection and restricted cohorts yielded consistent point estimates, these results should be interpreted cautiously. Overfitting remains a potential concern, and external validation in larger, independent cohorts is necessary before our findings can be considered definitive.

This study did not evaluate novel anti-CRAB agents such as cefiderocol or the sulbactam-durlobactam combination, which have shown promising in vitro and clinical activity against CRAB [26, 27]. During the study period, these agents were not yet routinely available at our institution, reflecting the current reality in most hospitals across China. Consequently, they were neither incorporated into treatment regimens nor included in routine antimicrobial susceptibility testing. Current Chinese guidelines continue to recommend polymyxin-based, sulbactam-based, and tigecycline-based combination therapies as the mainstay for CRAB infections [12], which formed the basis of our treatment strategies. As these novel agents become more widely accessible, future multicenter studies are urgently needed to evaluate their efficacy and safety in CRAB PCNSIs, and to determine whether they may offer superior outcomes or a more favorable safety profile compared to traditional polymyxin-based regimens. This evolving therapeutic landscape should be considered when interpreting our findings in the context of ongoing advances in CRAB management.

Conclusion

This study demonstrates that PCNSIs caused by CRAB are associated with significantly higher mortality than those caused by CSAB. In the absence of widespread availability of novel anti-CRAB agents, optimizing currently accessible therapeutic strategies remains critical. Our findings suggest that polymyxin B-based combination therapy, particularly when administered via routes that ensure adequate CSF penetration, may be associated with improved survival in CRAB PCNSIs. Additionally, timely post-infection drainage showed a protective effect in univariate analysis, suggestive of a potential benefit that warrants further investigation in larger prospective studies. In contrast, mechanical ventilation may represent a key risk marker, identifying patients who require heightened clinical vigilance. These results underscore the importance of integrating effective antimicrobial therapy with appropriate surgical management in this challenging infection. Future prospective, multi-center studies are needed to validate these findings and to evaluate the role of emerging therapies, including cefiderocol and sulbactam-durlobactam, in the evolving landscape of CRAB CNS infection treatment.

Electronic Supplementary Material

Below is the link to the electronic supplementary material.

Abbreviations

PCNSIs

Post-neurosurgical central nervous system infections

CRAB

Carbapenem-Resistant Acinetobacter baumannii

CSAB

Carbapenem-Susceptible Acinetobacter baumannii

CHINET

China Antimicrobial Surveillance Network

BBB

Blood-brain barrier

STROBE

Strengthening the Reporting of Observational Studies in Epidemiology

CSF

Cerebrospinal fluid

ICU

Intensive care unit

MICs

Minimum inhibitory concentrations

HIV

Human Immunodeficiency Virus

AIDS

Acquired Immune Deficiency Syndrome

CLSI

Clinical and Laboratory Standards Institute

EUCAST

European Committee on Antimicrobial Susceptibility Testing

IV

Intravenous

IT

Intrathecal

IVT

Intraventricular

OR

Odds Ratio

SD

Standard Deviation

LASSO

Least absolute shrinkage and selection operator

PCR

Polymerase chain reaction

mNGS

Metagenomic next-generation sequencing

Author contributions

Z.W. and P.Z. designed the research; P.Z. and C.L. accessed and verified the data, performed statistical analysis, and drafted the manuscript. All the authors participated in the interpretation of the results and critical revision of the manuscript.

Funding

Not applicable.

Data availability

Data is provided within the manuscript.

Declarations

Ethics approval and consent to participate

This retrospective study was approved by the Ethics Committee of Guangdong Sanjiu Brain Hospital (Approval No. 2016-Medical-01). The requirement for informed consent was waived by the same ethics committee in accordance with Article 39 of the Chinese National Health and Family Planning Commission regulation “Ethical Review of Biomedical Research Involving Humans” (2016). This study was conducted in accordance with the ethical principles of the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.