637 Clinical Prediction Model and Decision Support Tool for Ustekinumab in Crohn's Disease

INTRODUCTION: Identifying predictors of response to treatment with ustekinumab (UST) may improve therapeutic decisions. We created a clinical decision support tool (CDST) for UST therapy in active Crohn's disease (CD). METHODS: Patients from the UNITI trials were included if they received IV UST induction and SQ UST maintenance irrespective of Week 8 response status (derivation set; n = 781). Cox proportional hazard analyses were used to identify baseline factors associated with clinical remission (CDAI 150) by Week 16. The final model was transformed into a CDST and patients were stratified into 3 response probability groups (low, intermediate, high). Secondary analyses were performed to assess the ability of the CDST to predict exposure-efficacy relationships as measured by reduction in CDAI from baseline and differences in trough UST concentrations at weeks 8 and 16. RESULTS: In the derivation analysis, baseline albumin g/L (HR 1.041, 95% CI 1.019–1.063), no prior smoking history (HR 1.233, 95% CI 0.995–1.527), absence of baseline actively draining fistula (HR 1.330, 95% CI 0.906–1.952), absence of prior bowel surgery (HR 1.425, 95% CI 1.140–1.781), and absence of prior exposure to tumor necrosis factor antagonist (HR 1.591, 95% CI 1.214–1.900), were associated with achieving clinical remission by Week 16 of UST therapy. Rates of clinical remission at weeks 3, 6, 8, and 16 were significantly higher in the high probability group versus the intermediate and low probability groups (Table 1). The high probability group had a significantly greater reduction from baseline in CDAI compared to the intermediate and low probability groups at week 8 (high vs. intermediate 94 vs. 54 P < 0.0001; high vs. low 94 vs. 40, P < 0.0001) and week 16 (high vs. intermediate 155 vs. 112 P < 0.0001; high vs. low 155 vs. 84, P < 0.0001). The high probability group had significantly higher trough UST concentrations compared to the intermediate and low probability groups at week 8 (high vs. intermediate 5.2 vs. 3.6 μg/mL P = 0.052; high vs. low 5.2 vs. 2.2 μg/mL, P < 0.0001) and week 16 (high vs. intermediate 2.9 vs. 2.1 μg/mL P = 0.0053; high vs. low 2.9 vs. 1.2 μg/mL, P < 0.0001). CONCLUSION: Predictors of response to UST were identified and successfully modelled into a CDST which can stratify the probability of achieving early clinical remission and rapidity in onset of action in individual CD patients. A statistically significant relationship between UST trough concentrations, probability groups, and efficacy was observed.