Jaeyeon Kim, Juan Gonzalez Maffe, Eugene Tan, Janna Sand Dejmek, Claire Fabre, Christophe Meille
Novartis
Introduction
The Bayesian Logistic Regression Model (BLRM) is a well-defined statistical method for analyzing the relationship between dose and dose limiting toxicities (DLT) for dose escalation phase I studies. Semi-physiological PKPD models can describe longitudinal PD markers such as platelet counts. In this work, BLRM and PKPD models were used to support decision making during dose escalation and the prediction of maximum tolerated dose (MTD) for PCA062, an antibody drug conjugate (ADC) targeting P-Cadherin. PCA062 is an ADC containing the toxic payload emtansine (DM1), which has thrombocytopenia as an expected toxicity.
Objective
The objective of this work is to demonstrate how a PKPD model, developed for a specific toxicity (i.e., thrombocytopenia and/or platelet count decrease) may be combined with approaches such as BLRM, improve the decision making in a phase I dose escalation study when compared to only using DLT information alone.
Methods
PCA062 was administered by infusion (in mg/kg) every two weeks (Q2W). Ten dose levels ranging from 0.4mg/kg to 5.0mg/kg (Q2W) were tested during the dose escalation. Parallel evaluation of the BLRM and PKPD models’ predictions was conducted. The BLRM uses a two parameter logistic model to describe the dose-DLT relationship, where DLT is a binary outcome describing the occurrence of specific toxicities in patients (e.g., CTCAE grade 3 or higher). The prior distribution for the BLRM parameters was derived using preclinical data in monkeys. The probability of DLT and the risk that a given dose level exceeds the true MTD are estimated based on the posterior distribution of the model parameters. The safety PKPD model was introduced to support prediction of MTD after 29 patients had been treated in seven different dose levels. The PK model structure is a two compartmental model with linear elimination. The thrombocytopenia model from Bender et al. was adapted to describe platelet counts. The population thrombocytopenia PKPD model of PCA062 was developed from the clinical study data (n=29 , total number of platelets observations = 256). By using the estimated individual patient parameters, thrombocytopenia toxicity rates at different dose levels were simulated with the PKPD model. Estimation of parameters was done applying non-linear mixed effect modelling with Monolix 2016.
Results
Up to a dose level of 3.6mg/kg, both BLRM and PKPD modeling predicted a similar toxicity probability. Only the PKPD modeling approach accurately predicted the thrombocytopenia toxicity rates for unexplored dose ranges (>3.6mg/kg Q2W) suggesting 3.6mg/kg Q2W as a potential MTD for PCA062. However, when including all patient data up to 5.0mg/kg into the BLRM method both methods predicted comparable results. With the data from the 5mg/kg patients, the projected toxicity rates at 5mg/kg from the BLRM is 34.2% (95% CI 13.9 to 61.1), and the PKPD model is 31.7% (95% CI 18.1 to 48.1). Without the data from the 5mg/kg patients, the projected toxicity rates at 5mg/kg from the BLRM is 20.1% (95% CI 5.8 to 43.7), and the PKPD model is 37.8% (95% CI 20.5 to 57.7).
Conclusion
The BLRM and PKPD models were used during the decision to declare the MTD, with the PKPD model complementing the results from the BLRM. Using a semi-physiological PKPD model is relevant when the safety profile of a drug can be anticipated; in this case, PCA062 is anticipated to cause thrombocytopenia. While the BLRM incorporates all types of toxicities considered as DLTs as a binary outcome, the PKPD model describes only the platelet count observations, but as a longitudinal continuous measurement. Combining the BLRM model predictions with the a PKPD model can improve decision making in dose escalation studies. For the PCA062 phase I study, the PKPD model of thrombocytopenia was projecting a higher risk of thrombocytopenia DLT than overall risk of DLT from the BLRM method. Longitudinal measurement of a safety PD marker can be used to develop a PKPD model to supplement the BLRM results and have a significant impact on the decision making with respect to selecting an optimal dose for further development.
References:
[1] Bender BC, Quartino A, Li C, Chen S–C, Smitt M, Strasak A, Chernyukhin N, Wang B, Jin J, Girish S, Friberg LE. An integrated pharmacokinetic–pharmacodynamic (PKPD) modeling analysis of T-DM1–induced thrombocytopenia and hepatotoxicity in patients with HER2–positive metastatic breast cancer. PAGE Meeting 2016
[2] Neuenschwander, Branson, Gsponer. Critical aspects of the Bayesian approach to phase I cancer trials. Statistics in Medicine 2008
Reference: PAGE 28 (2019) Abstr 9108 [www.page-meeting.org/?abstract=9108]
Poster: Drug/Disease Modelling - Other Topics