Background: Fistula is a common, reoccurring complication of Crohn’s disease (CD). Accurate identification of fistula is essential to quantify burden and study treatment outcomes, particularly among patients initiating biology therapy given the widespread usage. In administrative data, diagnosis and procedure codes may be used to identify fistulizing CD. In electronic health record (EHR) data, natural language processing (NLP) of clinical notes may provide additional opportunity for identification.
Objectives: To identify and characterize fistulizing CD among patients initiating biologic therapy using structured codes and unstructured clinical notes
Methods: Adult CD patients treated with a biologic were identified within a large EHR and claims linked US database (Optum, Inc) from Jan 1, 2007 through Dec 31. 2019. Among patients with ≥2 years of EHR activity, use of biologic therapy was defined as a prescription claim, medication administration or procedure code in patients without prior use of another biologic (bio-naïve, n=5785) and with prior use of another biologic (bio-experienced, n=1783). Fistulizing disease was defined using ICD or CPT codes in the structured data and terms for fistula extracted from the unstructured clinical notes, ever prior to or on the date of biologic start. Characteristics were described for those with and without fistulizing CD.
Results: Approximately 12% (n=703) of bio-naïve patients had fistulizing CD identified using structured data. Inclusion of unstructured clinical notes increased the prevalence to 30% (n=1,713; 129 from structured data alone, 1,010 from unstructured clinical notes alone, and 574 with both). Among bio-experienced, the prevalence of fistulizing CD was 19% (n=342) when using structured data and 44% when unstructured clinical notes were included (n=785; 51 from structured data alone, 443 from unstructured clinical notes alone, 291 with both). Half (45% and 50%, respectively) of bio-naïve and bio-experienced patients identified by unstructured notes alone were prescribed antibiotics, namely metronidazole and ciprofloxacin. Compared to those without, patients with fistulizing CD were more likely to be younger, male, ever smokers, and have prior immunomodulator and steroid use.
Conclusions: The unstructured EHR identified a large proportion of patients with fistulizing CD, particularly those treated with antibiotics alone. Unstructured clinical text showed potential utility for identifying and characterizing patients with fistulizing CD. Further research is needed to assess if the disease burden or severity of fistula differs by data type.