Differentiating Viral and Bacterial Infections - A Machine Learning Model Based on Routine Blood Test Values

Aug 3, 2026

Upload by Himbo

Antibiotic resistance makes accurate differentiation between bacterial and viral infections essential for appropriate antibiotic use. This study develops a “Virus vs. Bacteria” machine learning model using 16 routine blood test results plus C-reactive protein, biological sex, and age. Using 44,120 cases from a single medical center, the model achieves 82.2% accuracy, a 0.129 Brier score, and an ROC AUC of 0.91. Performance exceeds traditional CRP decision rules, especially within the CRP range 10–40 mg/L where CRP alone is less informative. The approach supports improved clinical decision-making via multi-parameter biomarker tools.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages31

Reading Time1 h 18 min

AccessFree to download

Introduction

Antibiotic resistance and the need for accurate diagnosis

Common blood biomarkers: CBC, CRP, PCT and cytokines

Limitations of existing biomarkers and decision cutoffs

Role of machine learning in diagnostic medicine

Virus vs. Bacteria machine learning model

Input variables and dataset overview

Performance metrics and comparison to CRP rules

Differentiating Viral and Bacterial Infections: A Machine Learning Model Based on Routine Blood Test Values Gregor Gunčar,1,3\# Matjaž Kukar,1,2\# Tim Smole1, Sašo Moškon,1 Tomaž Vovko,4 Simon Podnar,5 Peter Černelč,1 Miran Brvar,6 Mateja Notar,1 Manca Köster1, Marjeta Tušek Jelenc1, Marko Notar*1 1Smart Blood Analytics Swiss SA, CH-8008 Zürich, Switzerland 2 Faculty of Computer and Information Science, University of Ljubljana, Slovenia 3 Faculty of Chemistry and Chemical Technology, University of Ljubljana, Slovenia 4 Department of Infectious Diseases, University Medical Centre Ljubljana, Slovenia 5 Division of Neurology, University Medical Centre Ljubljana, Slovenia 6Centre for Clinical Toxicology and Pharmacology, University Medical Centre Ljubljana, Slovenia \#Joint first authors contributed equally. *Correspondence: Marko Notar, [marko@sba-swiss.com](marko@sba-swiss.com) Abstract The growing threat of antibiotic resistance necessitates accurate differentiation between bacterial and viral infections for proper antibiotic administration. In this study, a Virus vs. Bacteria machine learning model was developed to discern between these infection types using 16 routine blood test results, Creactive protein levels, biological sex, and age. With a dataset of 44,120 cases from a single medical center, the Virus vs. Bacteria model demonstrated remarkable accuracy of 82.2%, a Brier score of 0. 129, and an area under the ROC curve of 0 .91, surpassing the performance of traditional CRP decision rule models. The model demonstrates substantially improved accuracy within the CRP range of 10-40 mg/L, an interval in which CRP alone offers limited diagnostic value for distinguishing between bacterial and viral infections. These findings underscore the importance of considering multiple blood parameters for diagnostic decision-making and suggest that the Virus vs. Bacteria model could contribute to the creation of innovative diagnostic tools. Such tools would harness machine learning and relevant biomarkers to support enhanced clinical decision-making in managing infections. Introduction The rise of antibiotic resistance poses a major threat to global public health, as it undermines the efficacy of life-saving antibiotics and increases the risk of complications and mortality associated with common infections (WHO, 2015) . A key driver of antibiotic resistance is the inappropriate use of antibiotics, particularly in situations where they are not clinically indicated, such as viral infections (Ventola, 2015) . Accurate and timely differentiation between bacterial and viral infections is essential to ensuring appropriate antibiotic prescribing practices and mitigating the spread of antibiotic resistance (Laxminarayan et al., 2013) . Healthcare providers commonly employ blood tests to obtain insights into a patient's health status. Complete blood count (CBC) and C-reactive protein (CRP) are among the most frequently measured blood test parameters in clinical practice, as they can provide valuable information about a patient's immune response and inflammation levels. The most commonly studied biomarkers for distinguishing between bacterial and viral infections include C-reactive protein (CRP), procalcitonin (PCT), and various cytokines (Chan et al., 2002; Hoeboer et al., 2015; van Houten et al., 2017) . Among these, PCT has shown the most promise due to its higher specificity and sensitivity in differentiating both bacterial infections from viral infections and bacterial infections from other noninfective causes of systemic inflammation (Simon et al., 2004) . PCT levels tend to be markedly elevated in bacterial infections, whereas they remain low in viral infections, providing a useful clinical tool to guide antibiotic therapy (Schuetz et al., 2017) . CRP, an acute-phase protein produced by the liver in response to inflammation, infection, or tissue injury (Pepys & Hirschfield, 2003), has also been widely used as a diagnos

FAQs

Why is differentiating viral and bacterial infections important in clinical practice?

Accurate differentiation enables appropriate antibiotic prescribing and helps limit the spread and impact of antibiotic resistance.

What inputs does the Virus vs. Bacteria machine learning model use?

The model uses 16 routine blood test results, C-reactive protein levels, biological sex, and age.

How does the model perform compared with traditional CRP decision rule models?

It reaches 82.2% accuracy with ROC AUC of 0.91 and improves outcomes over CRP-only rule approaches, particularly in the CRP range 10–40 mg/L.

Aug 3, 2026

Upload by Himbo