Machine learning reveals limited predictive value of clinical factors for asthma exacerbations

While predictors of asthma exacerbation risk are generally well established, predictors of exacerbation severity remain largely undefined. Identifying robust clinical predictors of exacerbation severity is essential to support tailored management strategies and optimize resource allocation. This study leverages machine learning to evaluate the predictive value of clinical factors for exacerbation severity in a real-world emergency department setting. A retrospective cohort study was performed using medical records of 367 adults (644 exacerbations) who presented to the Amsterdam UMC emergency department between 2013 and 2020. Five severity outcomes were investigated: hospital admission, ICU admission, length of stay, oxygenation efficiency (SpO₂/FiO₂), and National Early Warning Score (NEWS). Associations were assessed using linear mixed models (LMM), and predictive modelling employed a machine learning approach combining LMMs with fivefold cross-validated least absolute shrinkage and selection operator (LASSO) regression. Exacerbation severity was most consistently associated with lung function, the presence of a radiographic chest infiltrate, C-reactive protein levels, blood neutrophil count and theophylline maintenance use. No significant associations were found for blood eosinophil count, age, comorbidities, symptom duration, triggers, allergic sensitization, ethnicity or exacerbation history within the preceding 12 months. Internally validated prediction models for hospital and intensive care admission achieved areas under the curve of 0.632 and 0.695, respectively. The strongest predictors explained 18.8% of variability in NEWS, 15.2% in oxygenation efficiency, and 9.0% in length of hospital stay. In these prediction models, a radiographic chest infiltrate, followed by theophylline maintenance use and blood neutrophil count, were most frequently associated across the five severity outcomes. To conclude, lung function and markers of acute respiratory infection were most frequently associated with asthma exacerbation severity. However, clinical and demographic variables have only modest predictive value, highlighting the need to identify additional robust predictors.