ABSTRACT Purpose Given the increased likelihood for individuals with severe asthma to experience comorbidities, disease complications, emergency room visits, and hospitalizations, the ability to stratify asthma populations on severity is often important. Although pharmacoepidemiologic studies using administrative healthcare databases sometimes categorize asthma severity, there is no consensus on an approach. Methods Individuals with asthma (≥ 2 ICD‐10‐CM diagnosis codes J45) aged ≥ 6 years were identified in Optum's de‐identified Clinformatics Data Mart Database between January 2017 and November 2023. Severe asthma was inferred, consistent with the Global Initiative for Asthma (GINA), from prescription claims for high‐dose inhaled corticosteroids (ICS) in combination with long‐acting beta‐agonists (LABA) (Step 5 treatment). Two algorithm versions were employed to isolate the impact of dose estimation methods: (1) the “code‐based method” considered high‐dose ICS‐LABA to be an inhaler property and defined severe asthma based on claims for ICS‐LABA from our pre‐determined list; (2) the “calculation‐based method” considered high‐dose ICS‐LABA to be a regimen property and defined severe asthma based on derived patient‐level average daily dose. Results A total of 1 221 732 individuals with asthma were identified, 3.1% of which were severe by the code‐based method and 4.2% by the calculation‐based method. Both methods appeared to be consistent with the benchmark cited by GINA (3.7%). No meaningful differences were observed in the characteristics of the cohorts. 27% of calculation‐based individuals with severe asthma were not captured by the code‐based method. Conclusions Estimating patient‐level average daily ICS dose based on prescription claims using either a code‐based or a calculation‐based algorithm appears to be a reasonable method to identify individuals with severe asthma. The discrepancy between methods suggests that physician instructions sometimes vary from recommended administration instructions. Future work will validate these algorithms using electronic medical records.