Model-based safety thresholds for discrete adverse events
Tarjinder Sahota (1), Meindert Danhof (1), Oscar Della-Pasqua (1,2)
(1) Division of Pharmacology, LACDR, Leiden University, Leiden, The Netherlands; (2) Clinical Pharmacology Modelling & Simulation, GlaxoSmithKline, Stockley Park, UK
Objectives: Safety thresholds in preclinical toxicology are used to understand the risk/benefit profile of the drug and inform starting doses in first in man experiments. Traditionally, these are obtained with the No-Adverse-Event-Level (NOAEL) approach. It is an empirical method that calculates the minimum systemic exposure levels in animals with recorded adverse events. The discrete nature of many recorded AEs, and the possibility of unrecorded AEs, however, makes the assessment of these improvements technically challenging and also provides a challenge for the model-based methodology.
The objectives of this exercise are: a) to compare bias/accuracy of a model based approach to the NOAEL approach using "true" thresholds and b) demonstrate the feasibility of obtaining model-based safety estimates in standard toxicological experiments, for a variety of hypothetical AEs with different mechanisms.
Methods: An in-silico approach was used. Simulations were performed in NONMEM 6.2. The test species was rats and data was generated according to standard preclinical toxicological designs. A Markov model was used to simulate transition times and states representing the onset and severity of AEs. Four different biomarkers were used as covariates influencing transition probabilities:
Model 1) A one-compartment pharmacokinetic (PK) model with Michaelis-Menten elimination. Concentration of drug directly drives AE. Model 2) Irreversible binding to an enzyme which inhibits production of an endogenous compound with protective effects. Model 3) Formation of active metabolite with direct effects. Model 4) Indirect increase of production of a compound which drives AEs.
Safe-dosing thresholds were estimated via simulations and conservative safe estimates were obtained with simulations incorporating parameter uncertainty. Bias was assessed by deviation from a pre-defined safe exposure threshold and accuracy was also obtained and compared with the traditional NOAEL approach.
Results: The NOAEL approach exhibited the poorest accuracy and precision in comparison to the proposed methodology. The failure rate in determining a threshold was also higher.
Conclusions: Integrated use of data enables accurate modelling and predictive value. Even misspecified models outperform the NOAEL method. A model-based approach also allows for the incorporation of different definitions of acceptable risk for different AEs.