Towards understanding anti-infliximab antibody development to predict Crohn’s disease patients' underlying immunogenicity status
Alix Démaris (1,2), Elodie L. Plan (3), Wilhelm Huisinga (4), Linda B.S. Aulin (1), Sun Hee Lee (5), Joon Ho Lee (5), Walter Reinisch (6), Robin Michelet (1), Charlotte Kloft (1)
(1) Dept. of Clinical Pharmacy and Biochemistry, Institute of Pharmacy, Freie Universitaet Berlin, Germany, (2) Graduate Research Training program PharMetrX, Germany, (3) Department of Pharmacy, Uppsala University, Uppsala, Sweden, (4) Institute of Mathematics, University of Potsdam, Germany, (5) Celltrion, Inc, Incheon, Republic of Korea, (6) Dept. for Gastroenterology and Hepatology, Medical University of Vienna, Austria
Introduction: Inflammatory bowel diseases (IBD), including Crohn’s disease (CD), are a group of auto-immune diseases that affect the gastrointestinal tract. Anti-Tumour necrosis factor α (anti-TNFα) monoclonal antibodies (mAbs), such as infliximab (IFX), are efficacious in the treatment of mild to severe IBD. However, approximately one third of IBD patients receiving anti-TNFα mAbs do not respond to treatment, and up to half of them experience a loss of response (LOR) over time [1]. The development of anti-drug antibodies (ADA) in some patients has been related to low IFX concentrations through increased clearance [2], possibly leading to this LOR.
Occurrence of this immunogenicity depends on different drug-, patient-, and disease-related factors that, if identified, could enable to prevent the development of ADA [2,3] and thus help clinicians to anticipate the risk of LOR in those patients.
Objectives: The aim of this work was to predict the underlying immunogenicity status of CD patients receiving IFX therapy.
Methods: A randomised, double blind, multicentre, prospective phase 3 study, investigating a biosimilar of IFX developed by Celltrion, was conducted in 220 CD patients. All patients, naïve to mAbs, received an IFX standard dosing regimen (5 mg/kg administered at weeks 0, 2 and 6 and then every 8 weeks until week 54). ADA titers were measured via an ELISA immunoassay at weeks 0, 14, 30, 54 and at the end of study.
A mixed hidden Markov model (MHMM) was used to describe the underlying immunogenicity status, given the ADA titers measured; i.e. to describe the relationship between two stochastic processes: an unobserved, “hidden” process that follows a Markov chain, characterised by initial and transition probabilities, and an observed one, modelled as an independent process characterised by its distribution given the state [4]. This model was then used to predict patients’ most probable state at each time point, using the Viterbi algorithm [5]. This methodology has been successfully used in a previous study to characterise immunogenicity against Cimza®, another anti-TNFα agent [6].
Visual predictive check (VPC) and categorical VPC were performed for graphical evaluation of the models investigated. NONMEM 7.4.3 and R 3.6.0 were used for this analysis.
Results: A 2-state MHMM (states “No ADA producers” or S0, and “ADA producers” or S1) was developed. A zero-truncated Poisson (ZTP) distribution, characterised by parameter λ, was found to best describe ADA titers, considered as count data. λ of the ZTP distribution for ADA titers given the state S0 was fixed to a low value (λ0=0.0125) whereas λ describing ADA titers distribution given the state S1 was estimated close to the observed mean value (λ1=13.5, relative standard error (RSE)=15%).
The initial probability of being in state S0 was fixed to 99% as all patients were naïve to mAbs and thus should not be developing ADA at the start of the study. Estimated transitions probabilities were considered plausible with a probability of transitioning from S0 to S1 between two sample times of 17% (RSE=11%) and a lower probability of transitioning back to S0 of 10% (RSE=30%).
Interindividual variability (IIV) was estimated on all parameters, except the initial probability. A high IIV (>100%) was found for both λ0 and λ1, with a good precision (RSE<30%). Predicted immunogenicity status was in concordance with observed ADA status. Indeed for more than 90% of patients at each time points, the predicted state was the same as the one determined in the study.
Conclusions: Our 2-state MHMM properly characterised and predicted ADA development against IFX in our population. As a next step, this model could be extended to a 3-state model, by differentiating patients developing different types of ADA: Neutralising ADA (Nab) and Non-Nab.
A covariate analysis on this final model would then allow to identify factors impacting the probability of developing Nab or Non-Nab following IFX administration. Different dosing regimen could then be investigated, possibly preventing Nab or Non-Nab development, and thus decrease the risk of LOR in CD patients.
References:
[1] L. Peyrin-Biroulet, P. Deltenre, N. de Suray, J. Branche, W.J. Sandborn, J.F. Colombel. Efficacy and Safety of Tumor Necrosis Factor Antagonists in Crohn’s Disease: Meta-Analysis of Placebo-Controlled Trials. Clin. Gastroenterol. Hepatol. 6: 644–653 (2008).
[2] R. Keizer, A. Huitema. Clinical pharmacokinetics of therapeutic monoclonal antibodies. Clin. Pharmacokinet. 49: 493–507 (2010).
[3] J.T. Ryman, B. Meibohm. Pharmacokinetics of monoclonal antibodies. CPT Pharmacometrics Syst. Pharmacol. 6: 576–588 (2017).
[4] R.M.K. Altman. Mixed Hidden Markov models: An extension of the Hidden Markov model to the longitudinal data setting. J. Am. Stat. Assoc. 102: 201–210 (2007).
[5] A.J. Viterbi. Viterbi algorithm. Scholarpedia 4: 6246 (2009).
[6] A. Brekkan, B. Lacroix, R. Lledo-Garcia, S. Jönsson, M.O. Karlsson, E.L. Plan. Characterization of Anti-Drug Antibodies Using a Bivariate Mixed Hidden-Markov Model. PAGE 2018 (2018).
