Modeling the Dropout from Longitudinal Adverse Event Data: Selecting Optimal Titration Regimens
Bojan Lalovic, Matthew M Hutmacher, Bill Frame, Kaori Ito, Raymond Miller
Pfizer
Background: Dizziness and somnolence are the predominant adverse events (AEs) reported in the treatment of Generalized Anxiety Disorder (GAD) with pregabalin, representing a major determinant of study withdrawal (dropout). Various ad-hoc administration regimens (titration schemes) have been implemented throughout the development of this compound to reduce the incidence and severity of AEs and dropout rates, based on clinical considerations.
Objectives: To execute a quantitative, model-based titration schedule optimization, we modeled patient withdrawal resulting from reported AEs across all dose escalation regimens. The model is to be used to provide prospective clinical trial simulations of titration schedules to select an optimal scheme to minimize AEs and resulting study withdrawal.
Methods: We developed a parametric, discrete-time model of dropout (study withdrawal) based on individual daily self-reported AE (dizziness and somnolence) information. Adverse events were reported as none, mild, moderate or severe (ordered categorical data) across six 4-6 week studies. The dropout model focused on AEs due to considerable withdrawal during the initial, titration phase week. Graphics of nonparametric estimates of survival (plots of hazard probabilities vs. time and Kaplan Meier plots of observed vs. model predicted survivorship probabilities) were used as a guide in the selection of appropriate probability distribution model. Exponential, Weibull, log-logistic, linear-exponential and Gompertz parametric hazard probability distributions were examined to describe the time course of dropout risk as a function of the maximum adverse event by day. The hazard varied as a linear or exponential function of time for the linear-exponential and Gompertz models, respectively. The maximum adverse event severity (MAXAE) across dizziness or somnolence resulted in a more parsimonious model compared to the use of either measure of AE alone.
Results: The best fit to the data was achieved using the Gompertz model for hazard probabilities. This two term log-linear hazard function was comprised of time-invariant (intercept) and a time-dependent parameter (slope) allowing for a monotonically decreasing, time-dependent changes in discrete conditional dropout (hazard). Additionally, initial hazard was larger with higher severity of dizziness or somnolence (MAXAE>=2), corresponding to an initially short dropout half-life of 3, 8 and 60 days at day 1, 10 and 30, respectively. On the other hand, dropout half-lives were significantly longer (lower conditional risk) for groups reporting lower MAXAE severities, in the order of months and only slowly increasing throughout the study. For the individuals reporting no adverse events at day 1, 10 and 30, half-lives were 211, 250 and 350 respectively. The model also supported disease duration as a predictor of dropout. Predictive check demonstrated good model performance across all dose arms of the trials.
Conclusions: The discrete time parametric dropout model adequately described the time course of withdrawal across all GAD studies. Prospective simulations highlight the impact of differing titration schemes on dropout probabilities.
References:
[1] Feltner DE, Crockatt JG, Dubovsky SJ, et al. A randomized, double-blind, placebo-controlled, fixed-dose, multicenter study of pregabalin in patients with generalized anxiety disorder. J Clin Psychopharmacol 2003 Jun; 23 (3): 240-9
[2] Pohl RB, Feltner DE, Fieve RR, et al. Efficacy of pregabalin in the treatment of generalized anxiety disorder: double-blind, placebo-controlled comparison of BID versus TID dosing. J Clin Psychopharmacol 2005; 25: 151-8
[3] Rickels K, Pollack MH, Feltner DE, et al. Pregabalin for treatment of generalized anxiety disorder: a 4-week, multicenter, double-blind, placebo-controlled trial of pregabalin and alprazolam. Arch Gen Psychiatry 2005; 62: 1022-30
[4] Montgomery SA, Tobias K, Zornberg GL, et al. Efficacy and safety of pregabalin in the treatment of generalized anxiety disorder: a 6-week, multicenter, randomized, double-blind,placebo-controlled comparison of pregabalin and venlafaxine. J Clin Psychiatry 2006; 67: 771-82
[5] Pande AC, Crockatt JG, Feltner DE, et al. Pregabalin in generalized anxiety disorder: a placebo-controlled trial. Am J Psychiatry2003 Mar; 160 (3): 533-40
