Machine learning and clinician predictions of antibiotic resistance in Enterobacterales bloodstream infections

Aug 5, 2026

Upload by Sage

Active and timely antibiotic treatment for Gram-negative bloodstream infections is critical, yet identifying which patients harbor antimicrobial resistance (AMR) to guide empirical therapy remains difficult because microbiology results often arrive 24–48 hours later. This study builds XGBoost machine learning models to predict AMR to seven antibiotics in Enterobacterales bloodstream infection, using Oxfordshire hospital and community data from 2017–2021 and evaluating on 2022–2023 test sets and clinician prescribing.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

AccessFree to download

Article summary

Background

Methods

Findings

Conclusions

Journal of Infection 90 (2025) 106388 Contents lists available at ScienceDirect Journal of Infection journal [homepage:](homepage: www.elsevier.com/locate/j)[ www.elsevier.com/locate/j](homepage: www.elsevier.com/locate/j)inf | Infectious Disease Practice | | | | | --- | --- | --- | --- | | Machine learning and clinician predictions of antibiotic resistance in Enterobacterales bloodstream infections | | | | | Kevin Yuan a, Augustine Luk b, Jia Wei b, A. Sarah Walker b,c,d, Tingting Zhu e, David W. Eyre a,c,d,f,⁎
a Big Data Institute, Nuffield Department of Population Health, University of Oxford, Oxford, UK b Nuffield Department of Medicine, University of Oxford, Oxford, UK
c NIHR Health Protection Research Unit in Healthcare Associated Infections and Antimicrobial Resistance, University of Oxford, Oxford, UK dNIHR Oxford Biomedical Research Centre, Oxford, UK
e Institute of Biomedical Engineering, University of Oxford, Oxford, UK f Oxford University Hospitals NHS Foundation Trust, Oxford, UK | | | | | a r t i c l e i n f o | | s u m m a r y | | | Article history:
Accepted 20 December 2024 Available online 30 December 2024 | | Background: Patients with Gram-negative bloodstream infections are at risk of serious adverse outcomes without active treatment, but identifying who has antimicrobial resistance (AMR) to target empirical treatment is challenging.
Methods: We used XGBoost machine learning models to predict antimicrobial resistance to seven antibiotics in patients with Enterobacterales bloodstream infection. Models were trained using hospital and community data from Oxfordshire, UK, for patients with positive blood cultures between 01-January-2017 and 31-December-2021. Model performance was evaluated by comparing predictions to final microbiology results in test datasets from 01-January-2022 to 31-December-2023 and to clinicians’ prescribing. Findings: 4709 infection episodes were used for model training and evaluation; antibiotic resistance rates ranged from 7–67%. In held-out test data, resistance prediction performance was similar for the seven antibiotics (AUCs 0.680 [95%CI 0.641–0. 720] to 0.737 [0.674–0. 797]). Performance improved for most antibiotics when species identifications (available ∼24 h later) were included as model inputs (AUCs 0.723 [0.652–0. 791] to 0.827 [0.797–0.857]). In patients treated with a beta-lactam, clinician prescribing led to 70% receiving an active beta-lactam: 44% were over-treated (broader spectrum treatment than needed), 26% optimally-treated (narrowest spectrum active agent), and 30% under-treated (inactive beta-lactam). Model predictions without species data could have led to 79% of patients receiving an active beta-lactam: 45% over-treated, 34% optimally-treated, and 21% under-treated.
Conclusions: Predicting AMR in bloodstream infections is challenging for both clinicians and models. Despite modest performance, machine learning models could still increase the proportion of patients receiving active empirical treatment by up to 9% over current clinical practice in an environment prioritising antimicrobial stewardship.
© 2024 The Authors. Published by Elsevier Ltd on behalf of The British Infection Association. This is an open access article under the CC BY license ([http://creativecommons.org/licenses/by/4.0/](http://creativecommons.org/licenses/by/4.0/)). | | | Keywords:
Antibiotics Bloodstream infection Antimicrobial resistance Machine learning Antibiotic stewardship | | | | Introduction Active and timely antibiotic treatment of severe bacterial infections potentially saves lives and improves patient outcomes.1 However, it can take 24–48 h or more to obtain microbiology results to guide treatment, and many important infections remain culture ⁎ Correspondence to: Big Data Institute, University of Oxford, Old Road Campus, Oxford OX3 7LF, UK. E-mail address: [david.eyre@bdi.ox.ac.uk](david.eyre@bdi.ox.ac.uk) (D.W. Eyre). negative.

FAQs

Why is predicting antimicrobial resistance in Enterobacterales bloodstream infections challenging?

Patients can experience serious adverse outcomes without effective active treatment, but microbiology results take time and may not yet be available when prescribing decisions are needed. Additionally, translating resistance information into targeted empirical therapy is difficult for both clinicians and predictive models.

How were the machine learning models developed and evaluated in the study?

XGBoost models predicted resistance to seven antibiotics using data from Oxfordshire for positive blood cultures between 2017 and 2021. Performance was assessed by comparing predictions with final microbiology results in test datasets from 2022 to 2023 and against clinicians’ prescribing behavior.

What impact did including species identification have on prediction performance?

Model performance improved for most antibiotics when species identifications (available about 24 hours later) were included as inputs. Without species data, higher proportions of patients were predicted to receive active treatment, but misclassification could still lead to over- or under-treatment.

Aug 5, 2026

Upload by Sage