Background: Sublingual immunotherapies (SLITs) including Ragwitek (Short Ragweed Pollen Allergen Extract) were approved for marketed use in the US in April 2014. Because of the potential for serious allergic reaction, the first dose should be administered under physician supervision. SLIT initiation is likely to be missed in claims data as there are no relevant procedure codes, and the first dose may be provided as part of a sample pack. However, SLIT initiation may be observable in EHR data.
Objectives: To develop operational definitions to identify the initiation of Ragwitek within a large US EHR data system.
Methods: To identify potential Ragwitek users, we used natural language processing (NLP) to identify mentions of Ragwitek in the electronic clinical notes of the Humedica Research Database. De‐identified clinical notes were then sought and reviewed for a sample of patients to classify Ragwitek use (initiation, prevalent use, non‐use/insufficient information). Linking the note review results to the information extracted via NLP, we looked for patterns (NLP concept of Ragwitek in combination with NLP modifiers) that may indicate in‐office Ragwitek initiation.
Results: We identified 201 NLP concepts of Ragwitek from 133 unique notes among 95 patients in 2014. Of these, we requested 20 notes (among 19 patients); 19 notes (for 18 patients) were received and reviewed. The notes documented that 9 had initiated Ragwitek, 4 were prevalent users (i.e., initiation not captured by the note reviewed), and 5 were non‐users. Among the 9 initiators, 5 (56%) had an NLP concept of Ragwitek with a modifier of “administer” or “self‐administer”. Including mentions of an NLP concept of Ragwitek with a modifier of “start” and a recorded date corresponding to the date of the note increased the capture of Ragwitek initiators to 67%.
Conclusions: We identified 3 NLP concept algorithms that jointly captured 67% of the patients who initiated Ragwitek. Confirmation that these algorithms are generalizable to a broader sample of mentions of SLIT will be needed. Further refinement of the NLP patterns may also be needed.
