|
Abstract:
|
In scientific studies assessing the effect of one treatment over the other, much emphasis has often been placed of the "average" effect with less focus on the variability of the effects. Estimating the proportion of individuals who exhibit an effect opposite to the average may be important when evaluating treatment efficacy and safety. This paper presents an assessment of individual treatment effects for binary outcomes using the "Rubin Model of Causal Inference" which defines potential outcomes as (X,Y), where X and Y denote the response when a subject receives treatments X and Y, respectively, at a given time. The model defines a true individual treatment effect as D=X-Y. The analysis focuses on the TC-CT and TC-CT-TT-CC crossover designs. Using the first design, we construct bounds for the true "probability of negative effect." The second design helps estimate the same probability.
|
