Background: Evaluation of disease severity and treatment response in Crohn´s disease (CD) using Claims/EHR data is challenging due to the lack of disease specific lab or endoscopic results.
Objectives: To explore the feasibility of describing disease severity and treatment response approximation in Claims/EHR databases.
Methods: Patients prescribed >=1 anti-inflammatory biologic prescriptions: Adalimumab (Ada), Infliximab (Ifx), Certolizumab, Ustekinumab or Vedolizumab from 1 January 2016 to 30 June 2020 were extracted from Optum Clinformatics Data Mart (CDM) or Optum Market Clarity (MC) databases. Treatment failure was defined at the first switch to another biologic during the continuous biologic treatment episode of the index medication. Switches within the same medication classes or cross classes were identified. Covariates (comorbidities, medication, symptoms, encounter, lab tests, NLP notes) were screened to construct disease (Dx) severity profiles to estimate the reliability of the outcome. Descriptive analysis and chi-square test, ANOVA or cluster methods were used to show differences between failure and success groups.
Results: There were 11166 patients from 5 mutually exclusive biologics cohorts in CDM; 27275 patients in MC. Ada or Ifx groups represented most of the overall population in each database: 4200 (36.0%), 4398 (37.7%); 11,176 (41.0%) and 9,971 (36.6%), respectively. In CDM, 12.7% of patients were identified as treatment failures, including 3.3% of within-class and 8.4% across-class switch patients. In MC data, 17.5% patients were identified as treatment failures, including 5.8% of within-class and 11.7% across-class switch patients. To estimate other failure related events before proxy failure outcome, the frequency of taking rescue medication, specific failure related surgery during treatment episode or having hospitalized CD or CD complication/progression events during 60 days before the failure event or end of the follow up were assessed. The events rates were twice as high as the ones in success groups. Overall, in the failure group, 22.7% vs. 18.2% patients, as electronically confirmed switch patients, had relevant failure encounters before the switch by CDM or MC data.
Anemia, diarrhea, pain, medication usage of antibiotics or corticosteroids in addition to baseline switching history or concomitant medication class around index date were identified as important Dx severity profile factors. NLP symptoms or lab tests measure provided more information around switch event.
Conclusions: Using switch patients as a treatment response proxy for CD, the study results provide more understanding about which factors may be associated with switch event.
