Clinical trial simulations using a stroke disease progression model
Kristin E. Karlsson (1), Mats O. Karlsson (1) and E. Niclas Jonsson (1)(2)
(1) Department of Pharmaceutical Biosciences, Uppsala University, Sweden; (2) Exprimo NV, Mechelen, Belgium
Background: Combined categorical-continuous models such as the stroke disease progression models [1-3] are of particular use where data are relatively sparse, which is a typical scenario when analysing stroke scale data. Maximal use is made of the available information, and even missing observations can be informative. These models may provide significant advantages over current analytical methodology used in the interpretation of the score data routinely collected during stroke trials.
Objectives: To perform clinical trial simulations to calculate the power to detect a drug effect different from zero, using a disease progression model for NIH stroke scale (NIHSS) [2], and to assess the bias and precision of the drug parameter, under various conditions.
Methods: To be able to perform clinical trial simulations a drug effect parameter had to be introduced in the NIHSS disease progression model. Due to the complexity of the model, several options on where to introduce a drug parameter were available but initially the drug effect was added linearly on the magnitude of improvement. The dose-effect relation was calibrated such that a low, medium and high dose level would result in 25%, 33% and 50% fully recovered patients at end of study (the definition of a fully recovered patient was NIHSS<2 [4]). To assess the statistical power to detect a drug effect, multiple stochastic simulations and re-estimations (sse) were performed. Numerous datasets were simulated from the disease progression model with the added drug effect, and each dataset were estimated with two different models; with and without the drug effect (H1 and H0, respectively). To calculate the power to detect a drug parameter, a likelihood ratio test was performed on the difference in objective function value (outputted from NONMEM) between the H1 and the H0 models.
Results: Initial results, with 100 patients per dose arm, indicate a sufficient power to detect a drug effect parameter different from zero and that the bias and precision of the drug parameter are low. A rough assessment of the type-I error indicated that the nominal p-value for a 5% error rate was close to 0.05.
Conclusions: These initial results show the possibility to use a model based analysis within the stroke therapeutic area. This area has previously suffered from the requirement of very large trial sizes and a model based approach could enhance the feasibility of performing stroke trials.
References:
[1] Jonsson F, Marshall S, Krams M, Jonsson EN. A longitudinal model for non-monotonic clinical assessment scale data. Journal of Pharmacokinetics and Pharmacodynamics 2005; 32(5-6):795-815.
[2] K. Karlsson, J. Wilkins, M. Karlsson, N. Jonsson. Modelling disease progression in acute stroke using the NIH stroke scale [abstract]. AAPS annual meeting 2007. www.aapsj.org/abstracts/AM_2007/AAPS2007-002819.PDF
[3] KE Karlsson, J Wilkins, MO Karlsson and EN Jonsson. Modelling of disease progression in acute stroke by simultaneously using the NIH stroke scale, the Scandinavian stroke scale and the Barthel index [abstract]. PAGE 16 (2007) Abstr 1191 [www.page-meeting.org/?abstract=1191]
[4] The national institute of neurological disorders and stroke rt-PA stroke study group. Tissue plasminogen activator for acute ischemic stroke. N Engl J Med 1995; 333(24): 1581-1587.
