Pseudomonas aeruginosa-driven airway dysbiosis and machine learning prediction of acute exacerbations in non-cystic fibrosis bronchiectasis: a microbial-inflammatory signature approach

Abstract

Background: While Pseudomonas aeruginosa (PA) colonization is linked to poor outcomes in bronchiectasis, emerging evidence suggests that microbial community collapse—marked by diversity loss and depletion of commensal taxa—may better reflect disease progression than pathogen load alone. This study investigates whether airway microbiota dysbiosis driven by PA colonization induces ecological fragility and evaluates the predictive utility of integrating microbial diversity indices with systemic inflammation markers to forecast 1-year acute exacerbation risk using interpretable machine learning. Methods: Bronchoalveolar lavage fluid (BALF) samples from 23 patients (8 PA-colonized, 15 non-colonized) underwent 16 S rRNA gene sequencing. Microbial diversity and taxonomic composition were analyzed. An eXtreme Gradient Boosting (XGBoost) model with SHapley Additive exPlanations (SHAP) analysis was constructed to assess exacerbation risk, focusing on microbial and inflammatory markers. Results: PA-colonized patients (P1) exhibited significantly worse clinical severity than non-colonized patients (P2), with higher Bronchiectasis Severity Index scores (8.38 vs. 4.33, P < 0.01), poorer quality-of-life (SGRQ: 35.75 vs. 22.79; CAT: 24.00 vs. 16.26, P < 0.01), and elevated dyspnea (mMRC: 1.62 vs. 0.95, P < 0.05). P1 also had more acute exacerbations annually (retrospective: 3.00 vs. 1.20; prospective: 3.75 vs. 0.80, P < 0.05–0.001). Notably, P1 exhibited significantly reduced alpha diversity compared to P2 (Shannon index: 1.96 vs. 3.47; Simpson index: 0.46 vs. 0.77, P < 0.05). Weighted UniFrac PCoA revealed distinct clustering between groups (R²=0.162, P < 0.05). The XGBoost model, integrating microbial taxa relative abundances, alpha diversity indices, and inflammatory markers demonstrated robust performance in predicting 1-year acute exacerbation risk (AUC = 0.85). SHAP analysis identified the microbial diversity, rather than Pseudomona abundance was the most influential predictor of exacerbation risk. Conclusions: PA colonization disrupts airway microbial diversity and outcompetes commensal species in bronchiectasis, yet our XGBoost model reveals that ecological resilience—not pathogen load—best predicts exacerbation risk when integrated with inflammatory markers. This paradigm shift from pathogen-centric to ecosystem-driven risk assessment provides an actionable framework for personalized management and antibiotic stewardship in chronic airway diseases.