The Analysis of Efficacy Data of Drug Trials

Abstract

Typical efficacy endpoints have their associated statistical techniques. For example, values of continuous measurements (e.g., blood pressures) require the following statistical techniques:

• (a) if measurements are normally distributed: t-tests and associated confidence intervals to compare two mean values; analysis of variance (ANOVA) to compare three or more

• (b) if measurements have a non-normal distribution: Wilcoxon rank tests with confidence intervals for medians.

Comparing proportions of responders or proportions of survivors or patients with no events involves binomial rather than normal distributions and requires a completely different approach. It requires a chi-square test, or a more complex technique otherwise closely related to the simple chi-square test, e.g., Mantel Haenszl summary chi-square test, logrank test, Cox proportional hazard test etc. Although in clinical trials, particularly phase III–IV trials, proportions of responders and proportion of survivors is increasingly an efficacy endpoint, in many other trials proportions are used mainly for the purpose of assessing safety endpoints, while continuous measurements are used for assessing the main endpoints, mostly efficacy endpoints. We will, therefore, focus on statistically testing continuous measurements in this chapter and will deal with different aspects of statistically testing proportions in the next chapter.