In silico comparison of MTD determination in a phase I dose-finding framework
C. Vong (1,2), S. Fouliard (2), Q. Chalret du Rieu (2,4), I. Kloos (3), LE. Friberg (1), M.Chenel (2)
(1) Department of Pharmaceutical Biosciences, Uppsala University, Uppsala, Sweden, (2) Clinical Pharmacokinetic and Pharmacometrics Department and (3) Oncology business Unit, Institut de Recherches Internationales Servier, Suresnes, France, (4) Institut Claudius Regaud, Toulouse, France
Objectives: Determination of a maximum tolerated dose (MTD) still relies mainly on dose escalation studies using the algorithm-based 3+3 design although it has repeatedly been shown to be biased and to provide an imprecise MTD . Alternative methods, i.e. the Continuous Reassessment Method (CRM), the Escalation with Overdose Control (EWOC) and the Toxicity Probability Intervals (TPI), are increasingly gaining interest for Phase I studies. We propose here to develop an in silico CTS framework for multiple comparisons of MTD determination and to highlight the potential benefits to shift from a 3+3 design to more model-based methods.
Methods: Thrombocytopenia was used as the Dose Limiting Toxicity (DLT) in two 3+3 dose escalation trials in solid tumors (study 1) and in ovarian cancer (study 2). The “reference” recommended Phase II dose (RP2D) was assessed by simulating with the PKPD model [2,3] the dose resulting in 33% of the patients with grade 4 thrombocytopenia. A 3+3 design algorithm was implemented in R to compute the RP2D distributions. The CRM, EWOC and TPI were conducted using the bcrm R package , assuming a non-informative prior distribution and a power and two-parameter logistic functions to describe the dose-toxicity relationships. Stopping rule was stated when original study sizes were included.
Results: 120 and 135 mg/m2 were identified as the reference RP2Ds for study 1 and 2, respectively. The simulated distributions of RP2D 3+3 designs were centered at 75 mg/m2, with 90% CIs of 30-120mg/m2 for both studies. Less than 7% of simulated trials selected the reference RP2D and 91% of the trials underpredicted the true RP2D for study 1. Median RP2Ds were 105 mg/m2 using CRM and EWOC, and 90 mg/m2 using TPI. 37% of the patients in the 3+3 designs were treated at the therapeutic dose interval, unlike 45-54% of patients in the three other methods. 3+3 designs provided trials with less DLTs.
Conclusions: A comparison of different phase I oncology designs was successfully implemented using a thrombocytopenia PKPD model. The CRM method was shown to be in agreement with the reference model-based RP2D and provided a gain of 2 higher dose levels than the RP2D recommended from the 3+3 design approach. Furthermore, CRM, EWOC and TPI provided a better precision of RP2D. This work is in line with the methodology shift advocated by regulators and academics in phase I oncology studies.
 Iasonos Aand Ostrovnaya I. Estimating the dose-toxicity curve in completed phase I studies. Stat Med. 2011 Jul 30;30(17):2117-29.
 Friberg et al. Semiphysiological model for the time course of leukocytes after varying schedules of 5-fluorouracil in rats. J Pharmacol Exp Ther 2000 Nov;295(2):734-40.
 Friberg et al. Model of chemotherapy-induced myelosuppression with parameter consistency across drugs. J Clin Oncol 2002 Dec 15;20(24):4713-21.
 Sweeting et al. bcrm: Bayesian Continual Reassessment Method Designs for Phase I Dose-Finding Trials. Journal of Statistical Software (2013) 54: 1–26.