|
Abstract:
|
Randomized clinical trials are often used for evaluating treatment effect in evidence based medicine. Often due to distinct characteristics of patients, the effect of treatment on response may differ across subjects. In such cases, evaluating the effect of treatment on entire distribution of response may be of more interest instead of its mean. We propose an alternative bootstrap test with a counterintuitive method for comparing entire distribution between groups. This method includes student's t and rank t-test statistic for comparing means; first, second, and third quartile difference between groups for comparing first, second (median), and third quartiles respectively; and Kolmogorov-Smirnov test statistic for comparing different distributions between groups. For illustration, we utilized two randomized clinical trial datasets. The proposed bootstrap test provided the most comprehensive description and interesting treatment effects in both datasets which could not be captured by standard data analysis. The proposed approach may also be applicable for other designs where investigators might be interested in comparing between the two groups considering entire distributions.
|
