Integrating CBC-derived Inflammatory Biomarkers with Machine Learning for Discriminating Stable and Exacerbated COPD.

BACKGROUND: Chronic obstructive pulmonary disease (COPD) is a leading health burden and global cause of mortality. This study aimed to assess routine blood-derived biomarkers and integrate them with machine learning (ML) to improve detection and prognosis. METHODS: We conducted a retrospective analysis of 75 patients with stable and acute exacerbation COPD (AECOPD). CRP, ESR, WBC count, NLR, PLR, and LMR were compared among stable and AECOPD groups. Statistical analysis, fold-change, effect sizes, ROC-AUC, sensitivity analyses and correlation patterns were assessed. Demographic parameters including age and sex were also included in classifying biomarkers for AECOPD. Multiple ML classifiers were trained to predict disease state from biomarker patterns. RESULTS: NLR showed the most substantial elevation (Cohen's d =1.60) in AECOPD patients, followed by significant increases in CRP, PLR, and WBC. LMR significantly declined, while ESR's potential to discriminate AECOPD was non-significant. Feature importance analysis consistently ranked NLR, PLR, CRP and WBC as the top predictors, while ESR exhibited the least contribution. CONCLUSION: Integration of ML models with routine CBC-derived biomarkers, showed exploratory potential for distinguishing AECOPD from the stable state. Exacerbation remained the primary driver of inflammatory changes, irrespective of age and sex. Ensemble ML models outperformed traditional approaches.