Utilizing AI for decision support in severe asthma biological treatment

Background Severe asthma remains a major clinical challenge because of disease heterogeneity and variable response to biologic therapies. Despite the availability of clinical guidelines, clinicians often rely on a trial-and-error approach when selecting biologic treatments. We aimed to develop and validate an artificial intelligence (AI)-driven decision support system (DSS) to predict treatment response and support personalized biologic selection in severe asthma. Methods A retrospective multinational cohort of 801 patients with severe asthma from the United Arab Emirates, Saudi Arabia, Poland, and Canada was analyzed. Random Forest (RF) models were developed to predict clinical improvement, and Recursive Feature Elimination was used to identify the most informative predictors. Counterfactual simulation was subsequently applied to evaluate alternative biologic treatment strategies. Results The most important predictors of treatment response were Asthma Control Test (ACT) score, eosinophil count, and type 2 inflammatory comorbidities, particularly nasal polyposis and sinusitis. The optimized RF model improved specificity from 33.33% to 66.67% and increased overall accuracy from 78.57% to 82.14%. Counterfactual simulations demonstrated the model's ability to identify non-responder phenotypes and estimate alternative biologic options with higher predicted probabilities of improvement, including a 62% probability of improvement following a switch from anti-IgE therapy to anti-IL-5R therapy in selected patients. Conclusion This study demonstrates the potential of AI-driven clinical decision support to support personalized biologic selection in severe asthma. By improving identification of likely non-responders and simulating alternative treatment pathways, the proposed framework may help reduce ineffective treatment exposure, improve clinical decision-making, and advance precision medicine in severe asthma management.