Descriptive statistics.

Background
Acinetobacter baumanni infection is a leading cause of morbidity and mortality in the Intensive Care Unit (ICU). Early recognition of patients at risk for infection allows early proper treatment and is associated with improved outcomes. This study aimed to construct an innovative Machine Learning (ML) based prediction tool for Acinetobacter baumanni infection, among patients in the ICU, and to examine its robustness and predictive power.
Methods
For model development and internal validation, we used The Medical Information Mart for Intensive Care database (MIMIC) III data from 19,690 consecutive adult patients admitted between 2001 and 2012 at a Boston tertiary center ICU. For external validation, we used a different dataset from Rabin Medical Center (RMC, Israeli tertiary center) ICU, of 1,700 patients admitted between 2017 and 2021. After training on MIMIC cohorts, we adapted the algorithm from MIMIC to RMC and evaluated its discriminating power in terms of Area Under the Receiver Operating Curve (AUROC), sensitivity, specificity, Negative Predictive Value and Positive Predictive Value.
Results
The prediction model achieved AUROC = 0.624 (95% CI 0.604–0.647). The most significant predictors were (i) physiological parameters of cardio-respiratory function, such as carbon dioxide (CO2) levels and respiratory rate, (ii) metabolic disturbances such as lactate and acidosis (pH) and (iii) past administration of antibiotics.
Conclusions
Infection with Acinetobacter baumanni is more likely to occur in patients with respiratory failure and higher lactate levels, as well as patients who have used larger amounts of antibiotics. The accuracy of Acinetobacter prediction may be enhanced by future studies.