Skip to main content

Early prediction of carbapenem resistance phenotype in ICU-acquired Acinetobacter baumannii infection before susceptibility reporting: a temporally validated study

In brief

AI model predicts carbapenem-resistant Acinetobacter in ICU with AUC 0.88

Using routine ICU data, a machine-learning algorithm identified carbapenem-resistant Acinetobacter baumannii before susceptibility results, achieving an area under the ROC curve of 0.88 in a temporally separate test set. Key predictors included prior carbapenem use, low Glasgow Coma Scale, longer ventilation, and higher APACHE II scores. External validation is needed before clinical rollout.

Journal
International journal of antimicrobial agents (Q1)
Published
8 August 2026
Study design
Prospective / inception cohort
Evidence level
Level 2, Moderate (CEBM 2b)
Authors
Meng Zhou, Yaling Zhou, Chengxiang Xiang, Yiting Liu, Dongbo Li, Liru Zhao, et al.
PMID
42570778
DOI
10.1016/j.ijantimicag.2026.107957

Why clinicians should know about it

  • Picked for Microbiology (medical) (paper of the day, 11 August 2026): Early carbapenem resistance prediction model

Abstract

OBJECTIVES: To develop and temporally validate an interpretable approach using routinely available clinical data to differentiate carbapenem-resistant from carbapenem-susceptible Acinetobacter baumannii in intensive care unit (ICU) patients before susceptibility reporting. METHODS: This single-center retrospective cohort study included 423 critically ill patients with confirmed A. baumannii infection (June 2022 to December 2025), temporally split into training (n = 317) and test (n = 106) cohorts. Predictors were selected via univariate screening, least absolute shrinkage and selection operator (LASSO) regression, and multivariable logistic regression. A logistic regression nomogram and an eXtreme Gradient Boosting (XGBoost) model were evaluated for discrimination, calibration, and clinical utility using decision curve analysis and SHapley Additive exPlanations (SHAP) interpretability. RESULTS: In the temporal test cohort, the logistic regression model achieved an area under the receiver operating characteristic curve (ROC-AUC) of 0.831 and an area under the precision-recall curve (PR-AUC) of 0.916, while the XGBoost model achieved an ROC-AUC of 0.878 and a PR-AUC of 0.936. Prior carbapenem exposure, Glasgow Coma Scale score, duration of mechanical ventilation, and Acute Physiology and Chronic Health Evaluation II (APACHE II) score were the major contributors identified by SHAP analysis. Decision curve analysis demonstrated net benefit across clinically relevant threshold probabilities (approximately 0.10-0.80). CONCLUSIONS: This study developed and temporally validated interpretable models for early differentiation of carbapenem resistance phenotype to inform empirical antimicrobial therapy and antimicrobial stewardship before definitive susceptibility results. These findings require prospective multicenter external validation and clinical implementation studies before broader adoption in routine practice.

Abstract as published, via PubMed.

View on PubMedFull text at the publisherOpen in the app

For healthcare professionals. The summary is generated by AI from the published abstract, and the evidence level is assigned automatically from the study design on the Oxford CEBM hierarchy. Neither is medical advice. Read the full paper before changing practice.