A computational phenotype for difficult-to-treat asthma with corticosteroid overuse using electronic health care records.

Background: Patients with difficult-to-treat asthma experience insufficient symptom control despite intensive treatment, leading to overuse of oral corticosteroids (OCSs), which is associated with high patient and economic burden. Many of these remain unnoticed, even by their pulmonologist. Objective: We sought to create a query, also known as computational phenotype, identifying patients with difficult-to-treat asthma to facilitate their access to treatment optimalization. Methods: We used electronic health care records from 2 Dutch hospitals: Franciscus and Admiraal de Ruyter Hospital (Adrz) from the year 2023. We included adult patients with at least 1 hospital visit for asthma. The query was built using Structured Query Language code with data from Franciscus and subsequently applied to data in Franciscus and Adrz. Performance was scored by manual evaluation of the patient records by a pulmonologist and a pharmacist. Results: We included 1217 (Franciscus) and 267 (Adrz) patients. The query classified patients as having difficult-to-treat asthma when they had used high-dose inhaled corticosteroids during at least 90 days and had either received at least 2 courses of high-dose OCSs, or at least 90 days of low-dose OCSs in 2023. The sensitivity, specificity, positive predictive value, and negative predictive value were 0.75, 0.76, 0.63, and 0.85 in Franciscus, and 0.40, 0.95, 0.80, and 0.78 in Adrz, respectively. Conclusions: The query is an efficient way of identifying patients with difficult-to-treat asthma and OCS overuse.