Predictive Model for Outcomes in Inflammatory Bowel Disease Patients Receiving Maintenance Infliximab Therapy.

Abstract

Background: No models predict future outcomes in inflammatory bowel disease (IBD) patients receiving maintenance infliximab therapy. We created a predictive model for unfavorable outcomes.

Methods: Adult patients with IBD receiving maintenance infliximab therapy at 2 centers with matched serum infliximab concentrations and blinded histologic scores (Robarts Histopathologic Index [RHI]) were included. The primary endpoint was an unfavorable outcome of active objective inflammation or need for IBD-related surgery or hospitalization at 6-18 months follow-up. Internal variables were identified using univariable analyses, modeling used multivariable analysis, and performance was assessed (area under receiver-operating curve [AUC]) and externally validated.

Results: In 81 patients, 40.7% developed unfavorable outcomes at follow-up. Infliximab concentration <9.3 µg/mL (odds ratio [OR] 5.3, P = .001) and RHI > 12 (OR 3.4, P = .03) were the only factors associated with developing the primary unfavorable outcome. A prediction score assigning 1 point to each variable had good discrimination and performed similarly on internal (AUC 0.71) and external (AUC 0.73) cohorts. The risk of primary unfavorable outcomes in internal and external cohorts, respectively, was 23% and 15% for a score of 0, 46% and 50% for a score of 1, and 100% and 75% for a score of 2. Infliximab concentration alone performed similar to the 2-predictor model in internal (AUC 0.65, P = .5 vs. 2-predictor model) and external (AUC 0.70, P = .9, vs. 2-predictor model) cohorts.

Conclusions: Using unbiased variable selection, a 2-predictor model using infliximab concentrations and histology identified IBD patients on maintenance infliximab therapy at high risk of future unfavorable outcomes. For practical applicability, infliximab concentrations alone performed similarly well.