Mortality due to carbapenem-resistant Acinetobacter baumannii bacteraemia: a five-year cohort study in intensive care patients
Our latest research was published in the top-tier journal Clinical Microbiology and Infection (European Society of Clinical Microbiology and Infectious Diseases), on February 17, 2025.
Mortality due to carbapenem-resistant Acinetobacter baumannii bacteraemia: a five-year cohort study in intensive care patients
by
Stamatis Karakonstantis, Evangelos I. Kritsotakis, Renatos-Nikolaos Tziolos, Loukia Vassilopoulou, Maria Loukaki, Despoina Kypraiou, Emmanouil C. Petrakis, Alberto Tovil, Sophia Kokkini, Kyriaki Tryfinopoulou, Petros Ioannou, Εumorfia Kondili, Diamantis P. Kofteridis
Access the full paper via:
Abstract
Objectives:
Carbapenem-resistant Acinetobacter baumannii (CRAB) has emerged as a major and difficult-to-treat nosocomial pathogen. This study estimated the mortality associated with CRAB bacteraemia in patients receiving treatment in the intensive care unit. A susceptible-infection counterfactual framework was applied to reflect the potential benefit of improved antimicrobial therapy.
Methods:
A five-year (2019-2023) cohort study was conducted in a tertiary-care referral hospital in Greece. Competing risks survival analysis methods were applied to estimate excess in-hospital mortality due to CRAB bacteraemia by comparing patients infected by CRAB to those infected by other more susceptible Gram-negative bacteria (GNB).
Results:
The cohort comprised 400 intensive care patients with GNB bacteraemia (median age 70 years, 65% male). CRAB was the most common pathogen (43%), followed by K. pneumoniae (12%), E. coli (11%), and P. aeruginosa (10%). Patients with CRAB bacteraemia experienced significantly higher in-hospital mortality at 14 days (35% vs. 21%), 28 days (53% vs. 30%) and overall (74% vs. 52%) compared to patients with other GNB bacteraemia. Multivariable competing-risks regression confirmed that CRAB bacteraemia was independently associated with increased risk of 28-day inpatient death (cause-specific hazard ratio [csHR] 1.80, 95% CI 1.28–2.54; sub-distribution hazard ratio [sHR] 1.84, 95% CI 1.28–2.62), simultaneously lowering the probability of discharge alive (csHR 0.68, 95% CI 0.38–1.21; sHR 0.52, 95% CI 0.30–0.91). Estimation of the attributable fraction suggested that effective antimicrobial management may result in a relative decrease in the risk of in-hospital mortality by 44% (95% CI 22%–61%) in CRAB bacteraemia patients.
Conclusions:
CRAB's detrimental role as a leading cause of increased inpatient mortality and prolongation of hospitalisation in intensive-care patients was demonstrated. These outcomes could improve substantially if more effective antimicrobial treatment becomes available. Nevertheless, considering CRAB is predominantly a hospital-acquired pathogen, efforts should always be directed towards preventing nosocomial transmission.
Keywords: Carbapenem resistance; intensive care; bloodstream infection; Acinetobacter baumannii; survival analysis
Statistical methods
We calculated that 400 patients were required to detect moderately sized hazard ratios >1.5 (equivalently, <0.67) in a Cox proportional hazards model, with power 1–β = 90% and significance level α = 0.05, assuming 30% censoring and squared multiple-correlation coefficient of R² = 0.20 (equivalently, variance inflation factor of 1.25) when adjusting for covariates (Fig. S1).
Cumulative incidence function (CIF) plots were used to describe the risk of inpatient death, accounting for the length of follow-up and censored event times, and considering hospital discharge alive as a competing event that modifies the probability of inpatient death. The CIF gives the probability of a patient dying in the hospital at any given time, conditioned upon not experiencing either event (primary or competing) until that time. CIFs were estimated by applying the Aalen–Johansen method and compared between different groups of patients using the Pepe–Mori and Gray tests.
To account for variability in baseline covariates, cause-specific hazard ratios (csHR) were estimated for each event of interest (i.e. inpatient death and discharge alive) using multivariable Cox proportional hazards regression. The csHR is the instantaneous rate of occurrence of the event of interest (either inpatient death or discharge alive) on any given day in patients remaining hospitalized. Of note, a low csHR for discharge alive (<1) reflects a prolongation of LOS. The attributable risk fraction was calculated as 1–1/csHR to express the proportion of deaths occurring in patients with CRAB bacteraemia, which is in excess compared with the patients with bacteraemia by other GNB.
In the Cox model, the main exposure of interest was CRAB (vs. other GNB) bacteraemia. We adjusted for baseline confounders selected a priori by clinical judgement and experience, including age, sex, APACHE II score, CCI, prior surgery, pre-infection LOS, polymicrobial infection, PBS, and bacteraemia source. Multicollinearity was ruled out by examining variance inflation factors (Table S1). Log linearity of hazard rates for continuous covariates (e.g. age, APACHE II score, CCI, LOS, and PBS) was examined using Martingale residuals (Figs. S2 and S3) and restricted cubic splines were employed to model non-linear log hazards. The proportional hazards assumption for the effect of CRAB bacteraemia (i.e. constant effect over time) was confirmed by graphical assessments and the Grambsch–Therneau test (Fig. S4).
In sensitivity analyses, the robustness of the csHR estimates from the multivariable Cox model was examined under various restrictive conditions by excluding: (a) carbapenem-resistant isolates from the 'other GNB' group, (b) polymicrobial infections, (c) community-onset infections, and (d) all the previous combined. Moreover, the Fine-Gray model was applied to estimate sub-distribution hazard ratios (sHRs) in the full study sample and under the previous restrictive conditions to gain insights that align with cumulative incidence plots and complement the results from the cause-specific Cox models.
Time zero was set at bacteraemia onset. For the 14-day and 28-day outcomes, event-free time was administratively censored at 14 and 28 days, respectively, in patients who remained hospitalized for longer periods. There were no missing data for any study variable. Two-sided 95% CIs and p values were reported throughout. Statistical significance was considered when p < 0.05. Stata version 18 (Stata Corp., College Station, TX) was used. Commands stcompet and stpepemori were applied to estimate and compare CIFs.
Figure: Cumulative incidence plot of inpatient death and hospital discharge alive for N = 400 ICU patients with carbapenem-resistant A. baumannii (CRAB) bacteraemia versus other bacteraemia.
Figure: Cumulative incidence plot of inpatient death and hospital discharge alive for N = 400 ICU patients with bacteraemia caused by carbapenem-resistant A. baumannii (CRAB) versus bacteraemia by other pathogens.
Table: Sensitivity analyses of the effect of CRAB bacteraemia on the 28-day in-hospital outcome under various restrictive conditions
Figure: Sample sizes required for Cox proportional hazards regression to detect hazard ratios of inpatient death between 1.3 to 2.0 for the binary exposure of interest (CRAB bacteraemia), with power 1-β = 90%, significance level α = 0.05, assumed censoring rate of 30% , and assumed squared multiple-correlation coefficient R^2 =0.20 when adjusting for covariates (equivalent to a variance inflation factor of 1.25, i.e. sample size inflation by 25%). The method of Hsieh and Lavori (2000) was applied. The graph shows that with a sample size of about 400 patients, moderately sized hazard ratios >1.5 can be detected. [reference: Control Clin Trials 2000;21(6):552-60]