P333 Precise and unbiased infliximab dosing in patients with inflammatory bowel diseases using a multi-model averaging approach

Abstract

Abstract Background Underexposure to IFX is a common cause of loss of response in patients with inflammatory bowel disease (IBD). To ensure adequate – but not unnecessarily high – exposure, we aimed to identify a precise and unbiased approach for model-based dosing of IFX in patients with IBD. Methods A retrospective study was performed using data from, 54 patients on IFX maintenance therapy. The predictive performance of, 18 published IFX population pharmacokinetic (popPK) models was evaluated using NONMEM (v7.5). A priori prediction (only based on covariates) and Bayesian forecasting (BF; also based on one to three consecutively measured IFX trough concentrations; TC-2, TC-1, and TC0) of the IFX TC+1 was evaluated (Fig, 1). The predictive performance of a single-model approach was compared with two automated multi-model approaches: a model selection algorithm (MSA) and a model averaging algorithm (MAA).1 Relative bias (rBias) and relative root mean square error (rRMSE) were used to determine bias and imprecision of the predicted versus observed TC+1. Clinical acceptability was defined as an rBias between ±20% with a, 95%CI including zero. The predicted and observed TC+1 were classified at, 5.0 mg/L TC target.2 Results Four models were selected based on their predictive performances and implemented in the TDMx software tool to support model-based IFX dosing in the forthcoming prospective MODIFI study (NCT04982172). A priori prediction of TC+1 was clinically unacceptable with both single- and multi-model approaches (rBias +30% to +97%, rRMSE, 107% to, 158%; Fig, 2A). Also, a priori prediction had the lowest classification accuracy (median, 59%, IQR, 59%-63%; Fig, 3A). Providing one IFX TC greatly improved predictive performance (rBias -10% to +15%, rRMSE, 30% to, 49%; Fig, 2B, 2C) and classification accuracy (TC-1: median, 70%, IQR, 63%-72%; TC0: median, 80%, IQR, 76%-84 %; Fig, 3B, 3C). More specifically, BF resulted in a significantly lower chance of a falsely predicted ≥5 mg/L TC+1 than a priori prediction (p<0.01). In comparison with BF with TC-1, the availability of TC0 significantly lowered the chance of falsely predicting TC+1 <5 mg/L (unnecessary dose optimisation) (p=0.0034). Providing more than one previous TC improved predictive performances only marginally (data not shown). In general, MAA performed better than MSA. Conclusion Conclusion: A multi-model averaging approach provided more reliable Bayesian forecasts than the single-model approach. Adding one previous trough concentration in addition to covariate information sufficed to provide accurate and unbiased predictions of future exposure. Concentration data collected with a rapid assay may reduce the chance of performing unnecessary dose optimisation. Reference1. Uster D CPT,2021;2Vande Casteele N Gastroenterology,2017