The Use of Clinical Trial Simulation to investigate Bias in Crossover Studies with a Short Washout Period and no Placebo Arm: Application to Neuropathic Pain
G. Meno-tetang, A. Berges, S. Yang, and C. Chen
GlaxosmithKline, Greenford, UK
Objectives: Clinical trials in neuropathic pain are often run in parallel designs. The placebo response following administration of drugs such as gabapentin is now well characterized. This knowledge may be used to characterize the intrinsic response for doses not previously investigated. Although this correction with historical placebo data may be appealing when a parallel design is considered, a significant level of bias may be introduced in the case of a double blind 2-period crossover design. The placebo response of each patient in the second period may be influenced by the pain reduction perceived at the end of the first period. The objective of this work was to evaluate the level of bias that may be introduced in the comparison of the intrinsic effect of two doses in a study using a crossover design using historical placebo data obtained in the same indication and the same drug.
Methods: A parallel (reference) and a crossover trial with a short washout period were simulated with a low and a high dose. A pharmacokinetic-pharmacodynamic (PK/PD) model relating pain relief1 was used. The pain response in the first period of the crossover was deemed equivalent to that of a parallel design for the same dose level. However, drug response in the second period of the crossover was a function of the overall response (Placebo + drug) at the end of the first period, the level of patient expectation (number of points reduction in a visual analogue scale, VAS), and an arbitrary psychological factor.
Results: The results of the simulations show that in neuropathic trials using a crossover design, the placebo response in period two may be different from that of period one. If patients thought they were assigned to the low dose in the first period, they will experience a greater placebo response in the second period. Therefore, their expectations for the second part of the study will be raised. On the other hand, the placebo response in period 2 may be less important for subjects in the sequence: High Dose-to-Low Dose. The likelihood of seeing a difference between the two periods will be greater when the difference between the two dose sizes is large and when the level set for clinically significant response is smaller or equal to 2 points on a VAS. These simulations provide an approach to quantify the level of bias that may be introduced when data from a crossover study are corrected with historical placebo data obtained from a parallel design.
Conclusions: Given the subjective nature of the response to drugs in neuropathic pain, the placebo response may have a significant contribution to the overall response. Using historical placebo data from a parallel design may introduce a bias in the analysis of data from a crossover design even for the same drug and the same indication.
 P Lockwood et al. Pharmaceutical Research, Vol. 20, No. 11, November 2003