Machine learning prediction of 30-day mortality in adults with a positive blood culture
Enfermedades infecciosas y microbiologia clinica (English ed.) · 2026-11 · 44 (9) · 503265
DOI: 10.1016/j.eimce.2026.503265 ↗Kurz gefasst
Fachartikel in der Zeitschrift Enfermedades infecciosas y microbiologia clinica (English ed.), 2026-11. Verfasst von Ivan Maray und 2 weiteren Autor:innen. Beteiligte Einrichtung: Department of Pharmacy, Hospital Universitario Central de Asturias, Oviedo, Spain. Electronic address: ivanmaraymateos@gmail.com. Hauptthemen laut MEDLINE: Machine Learning, Blood Culture, Bacteremia. Die Arbeit nennt 23 Literaturangaben.
Abstract (Originalsprache)
INTRODUCTION: A positive blood culture is a clinically heterogeneous finding that may represent true bloodstream infection, transient bacteremia or contamination-diagnostic uncertainty that complicates early clinical decision-making. We aimed to develop and evaluate machine learning models to predict 30-day all-cause mortality in adults with a positive blood culture. METHODS: Retrospective cohort study using MIMIC-IV. Adult hospitalizations with at least one positive blood culture were included (first positive culture per admission); potential contaminants were retained to reflect real-world uncertainty. Demographic, clinical, laboratory, organ support and microbiological variables were used as predictors. Nine supervised machine learning algorithms were trained with repeated 10-fold cross-validation, evaluated in an independent test cohort and calibrated; thresholds were optimized with the Youden index. RESULTS: Of 5731 episodes, 742 patients (12.9%) died within 30 days. Test-set ROC AUC ranged from 0.759 to 0.801. After threshold optimization, the random forest model showed the best overall performance (sensitivity 0.788, specificity 0.707, balanced accuracy 0.748; Brier score 0.0999, calibration slope 1.03). ICU admission at onset, mechanical ventilation, bilirubin, urea and the Charlson comorbidity index were the main predictors. CONCLUSIONS: Machine learning models can provide clinically meaningful early prediction of 30-day mortality in patients with a positive blood culture using routinely available data. Focusing on the moment of blood culture positivity rather than established sepsis syndromes, this approach supports early risk stratification in a diagnostically heterogeneous population.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Department of Pharmacy, Hospital Universitario Central de Asturias, Oviedo, Spain. Electronic address: ivanmaraymateos@gmail.com.
- Research & Innovation, Artificial Intelligence and Statistical Department, Pragmatech AI Solutions, Oviedo, Spain.
- Research & Innovation, Artificial Intelligence and Statistical Department, Pragmatech AI Solutions, Oviedo, Spain; Departamento de Biología Funcional, Universidad de Oviedo, Oviedo, Spain.
Zitieren
APA
Maray, I., Valledor, P., & Fernández, J. (2026). Machine learning prediction of 30-day mortality in adults with a positive blood culture. Enfermedades infecciosas y microbiologia clinica (English ed.), 44(9), 503265. https://doi.org/10.1016/j.eimce.2026.503265
Vancouver
Maray I, Valledor P, Fernández J. Machine learning prediction of 30-day mortality in adults with a positive blood culture. Enfermedades infecciosas y microbiologia clinica (English ed.). 2026;44(9):503265. doi:10.1016/j.eimce.2026.503265
BibTeX
@article{Maray2026,
author = {Maray, Ivan and Valledor, Pablo and Fernández, Javier},
title = {Machine learning prediction of 30-day mortality in adults with a positive blood culture},
journal = {Enfermedades infecciosas y microbiologia clinica (English ed.)},
year = {2026},
volume = {44},
number = {9},
pages = {503265},
doi = {10.1016/j.eimce.2026.503265}
}
EUVIMED