Abstract
BACKGROUND: Accurate risk prediction of exacerbations is pivotal in severe asthma management. Multiple risk factors are at play, but the pathway of risk prediction remains unclear. RESEARCH QUESTION: How do the interplays of clinically relevant predictors lead to severe exacerbations in patients with severe asthma? STUDY DESIGN AND METHODS: Patients with severe asthma (n = 6,814, aged ≥ 18 years), biologic naive, were identified from the Severe Asthma Registry (2017-2021). Relevant predictors covered demographics, lung function, inflammation biomarkers, health care use, medications, exacerbation history, and comorbidities. A Bayesian network, representing the prediction process of severe exacerbations, was obtained by combining expert knowledge and machine learning algorithms. Internal validation was performed. The proposed influence diagram integrated decision and utility nodes into the prediction pathway. RESULTS: The Bayesian network analysis revealed that blood eosinophil count, fractional exhaled nitric oxide level, and FEV(1) directly influenced the transition between prior and future severe exacerbations. The presence of chronic rhinosinusitis indirectly affected such transition by directly influencing blood eosinophil count, fractional exhaled nitric oxide, and % predicted FEV(1). Macrolide use independently affected history of exacerbations to influence future severe asthma exacerbations. Model discrimination was moderate in 10-fold cross-validation and leave-1-country-out cross-validation, and model calibration was high in train-test data. INTERPRETATION: This study identified an essential prediction pathway of severe exacerbation, which involves the influence of chronic rhinosinusitis on the immediate predictors of risk transition from current to future severe asthma exacerbations. Macrolide use was another essential prediction pathway identified. The findings support shared clinical decision-making in severe asthma treatment.