PhD Scientific Days 2026

Poster Session 1.S - Conservative Medicine

Predicting bacterial aetiology in severe exacerbations of COPD using clinical biomarkers and machine learning

Name of the presenter

Sramkó, Bendegúz

Institute/workplace of the presenter

Semmelweis University, Department of Pulmonology

Authors

Bendegúz Sramkó¹, Balázs Csoma MD. PhD¹, Prof. Veronika Müller MD, PhD, DSc.¹, Zsófia Lázár, MD, PhD¹
1: Department of Pulmonology

Text of the abstract

Introduction:Bacterial infections are triggers of COPD exacerbations (COPD-AE), but access to microbiological verification is limited in clinical settings. While clinical features can signal aetiology, there remains a need for improved, data-driven tools to predict bacterial involvement at presentation to guide treatment.

Objectives: We aimed to investigate clinical characteristics and biomarkers at hospital admission in COPD-AE and their correlation with bacterial aetiology.

Methods: We analysed data from 89 patients in a previously published prospective cohort (Csoma et al., ERJ Open Res 2021). After defining bacterial aetiology based on positive sputum cultures, we compared these cases with those in which no pathogen was cultured. Evaluated variables included readily available blood biomarkers, such as CRP and the neutrophil-to-lymphocyte ratio (NLR), along with clinical history and spirometry. For predictive modelling, we used binomial logistic regression and a Random Forest (RF).

Results: We identified significant associations between bacterial aetiology and both age (p=0.01) and mid-high (5.4–10.8) NLR values (p=0.03). Our primary regression model (AIC=114.2) confirmed age (OR: 1.1; 95% CI: 1.02–1.15; p=0.01) and the mid-high NLR quartile (OR: 4.60 vs Q1; 95% CI: 1.13–18.6; p=0.03) as the strongest predictors of bacterial aetiology. Furthermore, we observed no association with current smoking (OR: 1.44; p=0.53) or the eosinophilic endotype (>0.3 G/L; OR: 0.37; p=0.16). Our RF model predicted bacterial aetiology with 70.1% accuracy (out-of-bag error: 29.9%) using age, NLR, and CRP.

Conclusion: Bacterial aetiology in COPD-AE is positively associated with advanced age and mid-high NLR values. Machine learning may aid the identification of bacterial COPD-AE.

Funding: This research was founded by the EKÖP grant (2025-363).