Artificial Intelligence for Predicting Response to Bronchodilator and Corticosteroid Therapy During Acute Asthma and COPD Exacerbations: A Systematic Review

Abstract

Background:

Artificial intelligence (AI) may support early identification of patients with acute asthma or chronic obstructive pulmonary disease (COPD) who are unlikely to respond adequately to bronchodilator and corticosteroid therapy, but its clinical utility remains uncertain.

Methods:

PubMed/MEDLINE was searched for human studies applying AI or machine-learning models to acute asthma or COPD exacerbations treated with bronchodilators and/or corticosteroids. Eleven cohort or prediction-model studies were synthesized narratively; no meta-analysis was performed because of substantial clinical and methodological heterogeneity.

Results:

Included studies ranged from 81 patients to 118,576 hospitalizations and evaluated treatment failure, hospitalization, critical-care disposition, readmission, and mortality. The most treatment-specific model predicted 30-day treatment failure in acute asthma/COPD with an area under the curve (AUC) of 0.81. In pediatric asthma treated with albuterol plus systemic corticosteroids,, while first-hour automated machine learning reached an AUC of 0.942. COPD readmission models generally showed lower discrimination, approximately 0.61–0.80, whereas 30-day mortality prediction reached an AUROC of 0.809 (95% CI 0.794–0.824). No study prospectively used AI to select drug, dose, or escalation strategy.

Conclusions:

AI shows clinically useful potential for early risk stratification during acute asthma and COPD care, particularly for hospitalization and deterioration. However, evidence remains insufficient for medication-specific response prediction; prospective, externally validated decision-support models are needed before routine clinical implementation.