Alienor Berges (1), Marc Cerou (1), Claire Ambery (1), Lia Liefaard (1), Stefano Zamuner (1), Emilie Hénin (2), Chao Chen (1)
(1) Clinical Pharmacology Modelling and Simulation, GSK, UK. ; (2) Université Claude Bernard Lyon1, Lyon, France ; UMR CNRS LBBE 5558 – équipe Evaluation et Modélisation des Effets thérapeutiques. Hospices Civils de Lyon, Service de Pharmacologie Clinique et Essais Thérapeutiques.
Objectives: The no-observed-adverse-effect level (NOAEL) of a drug defined from animal studies is crucial for inferring a maximal safe dose in human. However, several issues are associated with its concept, determination and application. For example, simulations showed that the NOAEL value which is identified for a given study highly depends on dose and sample size. [1-3]. We explored how modelling the probability of toxicity as a continuous function of time and dose across studies could potentially overcome these limitations.
The objective of this current work is to apply a TTE approach to histopathology toxicology data in rats for a test compound.
Methods: Binary histopathology data (presence/absence of toxicity, but not severity) from 7 studies (from 6 to 26 week study duration) were combined to develop a time-dose-toxicity model. The data were collected from each animal (n=137) once, at the time of terminal sacrifice. Parametric hazard modelling was conducted using the surv package in the R software (version 3.1.2) [4]. Model selection was mainly based on biological plausibility, AIC criteria, standard error values and simulation-based diagnostic plots. Covariates like dose level were tested in a univariate and multivariate manners.
Results: Due to the nature of data, the exact time of toxicity event was unknown, and all events were left- (toxicity appeared between the study start and sacrifice) or right-censored (toxicity had not appeared at the time of sacrifice). Hence time of observation was limited to a single time point per study, which resulted in 5 main time points in the total dataset. There were 6 dose groups with various sample sizes (from 9 rats in the lowest dose group to 136 in the control group). The model of choice was a time dependent hazard model with a weibull distribution and dose as significant covariate. The diagnostic plots showed a satisfactory fit of the data, despite the high degree of left censoring.
Conclusion: The diagnostic performance of the TTE model was similar to a previous logit model applied to the same data [3]. However, the TTE approach (including time as part of its definition) showed a more mechanistic description of the time impact on toxicity than the logit approach (including time as a covariate). Provided sufficient number of time points, the TTE approach would allow a better prediction of event incidence across study durations and would better inform the choice of experimental toxicity study designs.
References:
[1] U.S. Department of Health and Human Services, Food and Drug Administration, Center for Drug Evaluation and Research (CDER), Center for Biologics Evaluation and Research (CBER). Guidance for Industry, M3(R2) Nonclinical Safety Studies for the Conduct of Human Clinical Trials and Marketing Authorization for Pharmaceuticals, ICH. Revision 1. 2010
[2] Dorato MA, Engelhardt JA. The no-observed-adverse-effect-level in drug safety evaluations: Use, issues, and definition(s). Regul Toxicol Pharmacol. 2005;42:265-74
[3] Glatard A, Berges A, Ambery C, Osborne J, Smith R, Hénin E, Chen C. A model-based approach to support NOAEL determination: a simulation case illustrated by a real dataset, PAGE 2014
[4] David Diez. Survival analysis in R. http://www.stat.ucla.edu/david/teac/surv/R survival.pdf.
Reference: PAGE 24 () Abstr 3541 [www.page-meeting.org/?abstract=3541]
Poster: Drug/Disease modeling - Safety