Justin J Wilkins (1), Fredrik Jonsson (1), Per-Henrik Zingmark (2), Patrick Lyden (3), E Niclas Jonsson (1)
(1) Division of Pharmacokinetics and Drug Therapy, Department of Pharmaceutical Biosciences, Uppsala University, Uppsala, Sweden; (2) AstraZeneca LLP, Södertälje, Sweden; (3) Department of Neurology, Veterans Administration Medical Center, San Diego, California, USA
Objectives: This study was designed to refine analytical techniques for describing and predicting disease progression in acute stroke, by modelling scores measured on the Barthel Index, a categorical scale for assessing functional independence in recovering stroke patients, while addressing the statistical problem of dropout.
Methods: Scores assessed on four occasions over a 120-day period in 775 acute stroke patients were used to model the time course of recovery using NONMEM. Participating patients were recruited during the control arm of a double-blind, multinational, multicenter, placebo-controlled investigation of the efficacy of a novel acute stroke compound, At each measurement occasion, four discrete events were possible: attainment of a maximum score on the scale (‘healing’), improvement, decline, or dropout (the premature exit of a participant in the study). Each of these possible events had a probability and a score change magnitude associated with it. Scores were transformed – to constrain the predictions to the scale – and used to model the system. Time-related variables, including previous score magnitude and time elapsed since the previous observation, were considered as predictors, as well as physiological and demographic covariates such as age.
Results: To accommodate the non-monotonic nature of these transitions, it was necessary to develop a strategy that considered both the longitudinal, continuous aspects and the probabilistic characteristics of the data simultaneously. A basic framework incorporating both of these features was developed, and used to model the trajectory of disease progression using the scale scores. The final model’s predictive and descriptive abilities were good, and included covariate terms for previous score magnitude, time elapsed since previous observation, and age.
Conclusions: The model has a wide range of potential applications, including longitudinal analysis of stroke scale data, clinical trial simulation, and prognostic forecasting. This modelling approach is likely to be adaptable to other clinical assessment scales, such as those used in psychiatry and multiple sclerosis. Its use has great promise in reducing sample sizes and costs in clinical trials of drugs for the treatment of acute stroke, and thereby increasing the viability of research in this critical area.
Reference: PAGE 14 (2005) Abstr 744 [www.page-meeting.org/?abstract=744]
Poster: poster