Exposure-response modeling of daily seizure counts in focal epilepsy trials
Rik Schoemaker(1), Eric Snoeck(1), Armel Stockis(2)
(1) Exprimo, Mechelen, Belgium; (2) UCB Pharma, Braine-l’Alleud, Belgium
Objectives: To describe the change from baseline in individual seizure occurrences after treatment with brivaracetam or placebo in adjunctive treatment of focal epilepsy, and to establish an exposure-response relationship.
Methods: Individual daily seizure records were obtained from 1580 patients having participated in 5 brivaracetam double-blind, placebo-controlled trials: Four studies had a fixed dose parallel-group design and one had a flexible dose (optional titration) design. Individual plasma exposures were derived from a population PK model. Seizure probability was modeled on the log scale with NONMEM VI using statistical distributions appropriate for count data (Poisson, inverse binomial, with or without zero inflation, with or without Markovian feature) and was expressed as a function of baseline, placebo, dose or exposure, and subject-specific random effects.
Results: Inter-individual variability was implemented for baseline, placebo and Emax but not for ED50. Emax was allowed to take negative (improving patients) or positive values (deteriorating patients). The final model described the probability of seizure as a zero-inflated negative binomial distribution with between-patient variability on the zero inflation fraction and with Markovian feature. The latter allowed for a different seizure probability depending on whether the preceding day had seizures or not. Compared to the basic Poisson model, this model resulted in a drop of ~29,000 units in the Objective Function Value. Replacing the doses by individual exposures (AUCt or trough concentration) did not improve the goodness of fit.
The model was initially developed using the four fixed dose studies while the flexible dose study was used as external validation data set. Simulations showed that data of this type could be described well, even though the actual dose titration strategy based on clinical judgment could only be approximated. Analysis of the full data set including the flexible dose study resulted in model estimates that were clearly capable of describing the observed clinical endpoints, as demonstrated using visual predictive checks.
Conclusions: An exposure-response relationship was demonstrated with brivaracetam in patients with uncontrolled focal seizures. The daily count model provided superior goodness of fit due to its more detailed statistical structure and provided enhanced flexibility due to its ability to incorporate day-to-day changes in dose and covariates even in flexible dose designs.
Studies sponsored by UCB (N01114: NCT00175929; N01193: NCT00175825; N01252: NCT00490035; N01253: NCT00464269; N01254: NCT00504881).