Benjamin Guiastrennec (1), Anne Kümmel (1), Dongwoo Kang (2), Rajendra Singh (2)
(1) IntiQuan GmbH, Basel, Switzerland; (2) TEVA Pharmaceuticals, West Chester, PA, USA.
Objectives: Tardive dyskinesia (TD) is a serious movement disorder that appears as a complication of the prolonged use of dopamine receptor antagonists (e.g., antipsychotic agents). Austedo, a deuterated form of tetrabenazine, effectively reduces TD symptoms. In patients with TD, the Abnormal Involuntary Movement Scale (AIMS) measures the severity of abnormal movements, using a 5-point anchor. The AIMS Total Motor Score from Items 1 to 7 (i.e., orofacial, extremity, and truncal movements) is a composite score (range 0-28) representing the observed movements, with higher scores indicative of more severe dyskinesia. For a scale with such a large number of discrete levels, an ordered categorical modeling approach is not conceivable. Alternatively, treating the total score as a continuous variable on a logit scale ignores the categorical nature, and the extreme values would only be predicted asymptotically due to the logistic transformation. In this context, we evaluated the novel bounded integer (BI) modeling [1,2] as an alternative approach to predict changes in AIMS scores in response to twice daily administration of Austedo.
Methods: The model development was based on a pooled dataset from 2 Phase 2/3 studies comparing Austedo to placebo. Pharmacokinetic (PK) exposures parameters (i.e., average concentration (Cavg), maximal concentration (Cmax)) were predicted for the total concentration in the active Austedo metabolites (i.e., (α+β)-HTBZ) for each patient using previously developed and qualified population PK models. The individual total motor AIMS scores were modeled with the BI modeling approach in 2 distinct steps. At first, the model development focused on the description of the longitudinal placebo data, the evaluation of covariate effects on the baseline score (i.e., body weight, age, sex, race, and CYP2D6 genotype) and the assessment of a Markovian component. Different placebo effects were evaluated as such as constant (i.e., time independent), linear, or exponential decay over time on the mean of the latent variable distribution – a normal distribution was assumed for the latent variable. Then, data from the Austedo treatment arm was introduced, and the effect of the different PK exposure parameters was sequentially evaluated using linear and Michaelis-Menten kinetics on the mean of the latent variable distribution.
Results: In total, the modeling dataset included 1,964 total motor AIMS scores collected from 410 TD patients over the course of 12 weeks. The developed BI model was characterized by a linear effect of time (i.e., placebo), and by a linear effect of (α+β)-HTBZ Cmax on the baseline parameter of the latent variable distribution – nonlinear models were not identifiable. No covariate effect was included in the selected model.
A Markovian component implying that the current score depends on the value of the previous score was not supported by the model. The selected model appropriately predicted the total motor AIMS score, the change from total motor AIMS baseline score, as well as the proportion of each score over time. The predicted median change from baseline in total motor AIMS score was -1 point at week 12 in the placebo arm as caused by the time dependent effect. In the treatment arm the predicted median change from baseline in total motor AIMS score ranged from -1.6 to -3.4 points at 12 weeks for the first (0 – 24 ng/mL) and fourth (71 – 203 ng/mL) quartiles of (α+β)-HTBZ Cmax exposure, respectively.
Conclusions: The novel BI modeling approach was successfully applied to the prediction of the longitudinal exposure-response relationship of (α+β)-HTBZ in TD patients following Austedo administration. The developed exposure-response model could be applied to explore the predicted change in the AIMS score under various scenarios.
References:
[1] Wellhagen G.J, Kjellsson M.C, Karlsson M.O. A Bounded Integer Model for Rating and Composite Scale Data. The AAPS Journal 2019; 21:74.
[2] Ueckert S., Karlsson M.O, Improved numerical stability for the bounded integer model. JPKPD, 2020.
Reference: PAGE 30 (2022) Abstr 10018 [www.page-meeting.org/?abstract=10018]
Poster: Drug/Disease Modelling - CNS