Exposure-Response Analysis of Longitudinal Adverse Event Data.
Kaori Ito*, Matt Hutmacher#, Jing Liu, Ruolun Qiu, Bill Frame, Raymond Miller
Pfizer Global Research and Development (Groton, CT, USA) and A2PG (Ann Arbor, MI, USA)
Objectives: To describe the pregabalin-exposure and dizziness (adverse event) relationship in patients treated for Generalized Anxiety Disorder (GAD).
Methods: A model was developed for the incidence of dizziness (adverse event) and for the conditional severity of dizziness in patients that have at least one incidence over the duration of the study. Dizziness was recorded on a four point categorical scale (0=none, 1=mild, 2=moderate, 3=severe) for 6 clinical studies in patients with GAD. The unconditional dizziness severity was obtained by calculating the joint probability of the incidence and severity of dizziness. The incidence component was modeled using a nonlinear logistic regression model. The conditional severity component was modeled as an ordered categorical variable with proportional odds. The exposure response relationship was evaluated as a linear or Emax relationship. To describe the time-course of severity, time-dependent effects (placebo and tolerance) were also included. A Markov element was introduced to account for the correlation between adjacent observations. To evaluate the predictive properties of the model, a posterior predictive check was performed. One hundred data sets were simulated from the final conditional severity model with and without the Markov element and the number of transitions between each possible transition were calculated.
Results: The dataset prepared for the 6 studies consisted of 47218 observations collected in 1630 patients. For the incidence model, a sigmoid Emax model best describes the dose-dizziness response relationship. For conditional severity, the model that best described the data was an Emax model with a placebo time-course response and a component that allows for an exponential attenuation of the dizziness severity. The numbers of observed transitions for all combinations of dizziness severity were contained within the predictive check distributions from the Markov model, while the number of transitions were markedly overestimated or underestimated without the Markov element.
Conclusions: The probability of experiencing dizziness during any day increases with pregabalin daily dose. The predicted mean incidence of dizziness was around 35 % at a daily dose of 200 mg/day or greater, which was at least 2 fold higher compared to those at daily doses <150 mg/day. The most frequently reported severity was mild to moderate. The risk of mild or moderate dizziness increases up to 25 % within 1 week, but declines to around 7 % over 3 to 4 weeks. The proportional odds model including a time course of appearance and disappearance of adverse event could adequately describe the time-course of probability of dizziness. Incorporating a transition model including Markov elements improved the model fit and greatly improved the predictability of the time-course of probability of dizziness.
