Predictive models assessing the response to ustekinumab highlight the value of therapeutic drug monitoring in Crohn's disease

Abstract

Background: Despite the therapeutic efficacy of Ustekinumab (UST) in Crohn's disease (CD), loss of response (LOR) is observed over time. This study aims to evaluate the impact of the UST pharmacokinetics (PK) at induction on clinical and endoscopic outcomes, as well as to find predictive markers of UST response.

Methods: This retrospective study included 80 CD patients. Pharmacokinetics data (trough levels (TLs)) combined with clinical and biological parameters were fed into tailored logistic regression and tree-based ensemble techniques to predict clinical and endoscopic outcomes at one year of follow-up.

Results: TLs at week 16 were significantly lower among patients with moderate to severe endoscopic activity during the follow-up (p = 0.04). The best model to predict endoscopic outcome was obtained at week 16 by Random Forest with an area under the receiver operating characteristic curve of 0.92 ± 0.08, sensitivity 91% and specificity 75%, with key inputs such as lymphocyte and monocyte counts at week 8, and UST TLs and CRP at week 16.

Conclusions: This real-world study confirms the relationship between early UST TLs and both clinical and endoscopic outcomes. Models were developed for the task of predicting clinical and endoscopic remission in CD patients treated with UST, highlighting the clinical relevance of UST TLs at week 16.

Keywords: Crohn's disease; Pharmacokinetics; Predictive models; Therapeutic drug monitoring; Ustekinumab.