Population methods for dose escalation studies: an MCMC approach
A. Russu (1), M. Neve (2), G. De Nicolao (1), I. Poggesi (2), R. Gomeni (2)
(1) Dipartimento di Informatica e Sistemistica, Università di Pavia, Italy; (2) GlaxoSmithKline, CPK, Modelling & Simulation, Verona, Italy.
Objectives: In dose escalation studies decisions must be taken on the next doses to be administered based on safety and PK outcomes. Recently, there has been a growing interest in Bayesian methods for the sequential estimation of safety risk [1] and dose-exposure relationship [2, 3]. The approach proposed in [2], which relies on a standard linear mixed-effect model relating systemic exposure to dose levels after logarithmic transformation of the data, assumes a fixed dependence of interindividual variability on measurement noise and constrains all the individual curves to have the same slope on log-log scale. The aim of the present work is to develop a more general Bayesian method which overcomes these limitations.
Methods: The removal of the simplifying assumptions made in [2] hinders the closed-form derivation of the posterior distributions. Therefore, it was necessary to resort to a MCMC implementation of the estimation procedure using WinBUGS, which provides posterior distributions of the systemic exposures (Cmax, AUC) at new doses of both tested and untested subjects.
Results: The proposed approach was validated on both simulated and clinical data. Different datasets were simulated, changing both sample size and number of assessments/subject. The estimated dose-exposure relationships were compared with the true ones so as to assess how the experimental design affects the reliability of the estimates. Compared to [2], the MCMC approach provides a more realistic assessment of the estimation uncertainty. The clinical study considered was a single-blind, randomised, placebo-controlled design in 2 cohorts of 10 healthy male subjects to assess safety, tolerability and PK of single ascending doses of a new candidate drug. Subjects of both cohorts underwent 5 treatment sessions (placebo and 4 ascending doses). PK results of each treatment session were timely evaluated so as to avoid that individuals exceeded predefined limits, based on threshold values of Cmax and AUC.
Conclusions: The proposed method offers a flexible and general framework supplying an effective aid to the clinician. Future developments may regard its integration within a decision support system for sequential optimal dose selection.
References:
[1] O'Quigley J, Pepe M, Fisher L (1990). Biometrics 46: 33-48.
[2] Whitehead J, Zhou Y, Patterson S, Webber D, Francis S (2001). Biostatistics 2: 47-61.
[3] Berry DA, Müller P, Grieve AP, Smith M, Parke T, Blazek R et al. In Gatsonis C, Kass RE, Carlin B, Carriquiry A, Gelman A, Verdinelli I, West M (eds.), Case Studies in Bayesian Statistics V. New York: Springer-Verlag; 2001: 99-181.
