Sven van Dijkman (1), Meindert Danhof (1), Oscar Della Pasqua (12)
(1) Division of Pharmacology, Leiden/Amsterdam Center, Leiden University, The Netherlands; (2) GlaxoSmithKline London, UK
Objectives: Differences in response rate and efficacy occur across patients and across epilepsy types. Such differences may be explained by intrinsic (e.g., insensitivity to mechanism of action (MoA) ) and extrinsic factors (e.g., endpoints, measurement method). The possibility of predicting treatment response taking into account pharmacological effects, placebo response and disease progression is critical for drug development, as the effect of add-on therapy can be confounded by the underlying disease progression. Despite the high incidence of epilepsy in the paediatric population, no systematic efforts have been made to describe its course and progression.
Methods: Given the chronic and paroxysmal nature of seizures, epilepsy can be described by Markovian processes. Markov models consist of series of states and transition rates which reflect the course of the disease, transitions between states can be used to describe placebo (PE) and treatment effects (TE). Here we propose the use of a hidden markov model (HMM) to relate frequency of seizures to underlying activity. PK [1] and PD profiles for valproate (VPA)-treated paediatric epilepsy patients were simulated using literature data. A HMM was fitted to these data describing short-term (ictal vs. interIctal) and long term (remission, resistance and death) states, including parameterisation of TE (Emax model) on the transition rates. The feasibility of this approach to discriminate between drug and disease-specific parameters and the stability of the model were evaluated using a bootstrap procedure.
Results: Five states and their respective transitions were implemented in the HMM. The probabilities of the transitions of individuals between states were modeled as the sum of the natural course (NC), the PE and TE. Parameters estimated using the Laplacian method were NC, PE, onset of PE and KE0. The model described the gradual decrease in seizure frequency including a more profound drop between the 5th and 10th day, representing an effect delay due to MoA and loading phase of the drug. Diagnostics showed adequate goodness-of-fit, with clear distinction between TE, PE and NC of disease.
Conclusions: Preliminary results suggest that a HMM can be used to accurately fit longitudinal data from chronically treated patients. Although demographic covariate effects have not been implemented yet, transition rates were shown to be sensitive to drug effects and may provide the basis for predicting long term response in the presence of disease progression.
References:
[1] Valproate population pharmacokinetics in children, B. Blanco Serrano, M.J. GarcÃa Sánchez, M.J. Otero, D. Santos Buelga, J. Serrano, A. DomÃnguez-Gil. Journal of Clinical Pharmacy and Therapeutics (1999) 24, 73-80.
Reference: PAGE 21 (2012) Abstr 2566 [www.page-meeting.org/?abstract=2566]
Poster: Paediatrics