Late Breaking Abstract - iPREDICT: Pilot study to develop predictive model of changes in asthma control

Abstract

Background: iPREDICT aimed to predict asthma control changes using digital sensor technology (Hagger L et al., AJRCCM 2019). Objectives: With data on common, noninvasive biomarkers, we evaluated algorithms to identify patient-specific triggers and predict changes in asthma control. Methods: In a 24-wk pilot study, 108 US pts with severe asthma interacted with 3 sensors, 1–2 connected devices, and 2 mobile applications. An asthma event (endpoint) was defined as asthma symptom worsening (either logged by pts, PEF 8 SABA puffs in 24h, or >4 puffs/day over 48h). For each endpoint, predictive models were built at population, subgroup, and individual levels (fig). Results: Predictive accuracy depended on endpoint selection. Because of disease heterogeneity, population-level models inaccurately predicted asthma endpoints. Subgroup models (most accurate predictive endpoint: >4 puffs/day over 48h) had relatively high accuracy for certain groups (fig). Individual models were reasonably accurate, with sufficient, good-quality data. Analysis of high-accuracy individual models indicated important predictive biomarkers for each endpoint, which differed greatly between pts. Conclusion: We demonstrated the feasibility of developing subgroup and individual predictive models in asthma. These models, continuously trained on pts’ own data, could potentially predict asthma events with relative accuracy.