C25-19 Leveraging Electronic Health Record Data to Predict Response to Biologics in Severe Childhood Asthma

Abstract Background Treatment with monoclonal antibodies (“biologics”) can substantially reduce the risk of severe asthma exacerbations (SAEs) among children with moderate or severe asthma. However, not all patients who initiate biologic treatment experience such improvement. Here, we use clinical data that is routinely collected in the electronic health record (EHR) to predict biologic response. Methods We analyzed longitudinal EHR data from children with moderate/severe asthma treated with a biologic by pediatric subspecialists at our center between December 2019 and December 2024. We identified clinical correlates of increased SAE risk at baseline (prior to biologic start) and used logistic and Cox proportional hazards models to identify and quantify their prognostic utility for biologic response. Our main outcome was any SAE in the year following biologic initiation. Results We identified a cohort of 122 children (51% female, 61% Black, 4% Hispanic) with moderate/severe asthma on biologics (63% dupilumab, 27% omalizumab, 10% mepolizumab), including 15.6% who had ≥1 SAE (and 13.1% who had ≥2 SAEs). Omalizumab and mepolizumab had shorter times to SAE compared to dupilumab (p 0.01). Biologics had differential effects on the risk of SAE depending on a child’s sex (interaction-p=0.024), age at treatment initiation (interaction-p=0.01), lung function (% predicted FEF25-75: interaction-p=0.034), and absolute neutrophil counts (interaction-p=0.055) (Figure 1). Dupilumab was associated with lower risk of SAE among male patients; and omalizumab and mepolizumab were associated with higher risk of SAEs among patients with high absolute neutrophils and younger age at treatment initiation. Consideration of these clinical parameters in a multivariable model resulted in 86% accuracy to predict SAE after starting biologics (0.86, 95%CI=0.79-0.89), a 23% improvement compared to reliance on previous history of SAEs alone (0.63; 95%CI=0.54-0.72). Conclusion Routinely collected clinical parameters could help improve prognostic accuracy for biologic response and aid clinicians in tailoring therapy. Further studies will be needed to validate these findings. This abstract is funded by: NIH