DISEASE PROGRESSION PHARMACOKINETIC MODELING IMPROVES THE ACCURACY OF EARLY INFLIXIMAB CONCENTRATION TARGETS

Background: Our team (Xiong et al. 2021) recently developed an infliximab population pharmacokinetic (popPK) model in children and young adults with Crohn’s disease (CD). We found that prediction accuracy improved when all five covariates of drug clearance were included. Given the dynamic changes in the four covariates (weight, albumin, sedimentation rate, neutrophil CD64 expression) that occur during induction, we hypothesized that model-informed precision dosing to predict early trough concentrations (cTrough) from baseline covariates alone would be imprecise. Therefore, our aims were to: 1. Test the precision of the model predicted (a priori) cTrough compared to the observed cTrough at dose3 and dose4 and 2. Test whether disease progression modeling would provide more precise predictions.
Methods: REFINE is a prospective cohort of 78 children and young adults receiving infliximab at four medical centers from 2014-2019. The DSH retrospective cohort includes 161 children and young adults with CD who started infliximab at a single center from 2019-2021. REFINE cTroughs were obtained at every infusion while the DSH cohort had cTroughs obtained either at dose3, dose4 or both. Model imprecision was calculated by the root mean square error (RMSE%) and percent bias by the mean prediction error (MPE%). The Pearson r and R² were used to calculate the correlation between the observed and the model predicted cTrough.
Results: The median (IQR) age at the start of infliximab was 14.3 (11-16) years with 32% female, and 85% white race. The median starting dose was 7.1 (5.5-10) mg/kg. Using the combined cohorts, we first calculated the observed percent improvement of all four covariates from dose1-dose3 and from dose1-dose4 (Table1). We then assessed the a priori predicted cTrough at dose3 and dose4 for both cohorts using the Xiong et al. popPK model and Bayesian estimation under three conditions. Condition1: cTrough were predicted using only the four baseline (pre-treatment) covariates of clearance. Condition2: cTrough were calculated using a combination of the baseline covariates and the calculated improvement (from baseline as shown in Table1) in all four covariates at either dose3 or dose4. Condition3: combination of Condition2 and one observed cTrough (week2 for dose3 and week6 for dose4 predictions). As shown, the MPE% improved when the simulated dose3 covariates were included in the prediction (Table2, A2). Furthermore, the correlations between the predicted and observed dose3 cTrough also improved for ConditionA3. For dose4, incorporating the week6 level improved the RMSE% and Pearson correlation (ConditionB3).
Conclusions: Disease progression modeling improves the precision of predicting the targeted dose3 (week6) cTrough. Disease progression modeling will be incorporated in our PK dashboard to perform more accurate model-informed precision dosing.