ABSTRACT
Chapter 9 is a treatise on monitoring clinical trials using a unified Brownian motion (‘B-value’) approach. Many different test statistics can be transformed to B-values, allowing the same boundaries that apply to a t-test to be valid for all of these test statistics. All that is needed is a measure of the amount of information available at interim analyses and final analyses. We define information and information fraction and give a heuristic interpretation. Historic boundaries such as those of Pocock (1977), O'Brien and Fleming (1979), and Haybittle (1971) are included. We emphasize the Lan and DeMets (1983) alpha spending function approach, which does not require interim analyses to be equally-spaced in terms of information. We pivot to futility monitoring via conditional power, stochastic curtailment, predicted interval plots, and beta spending functions. We illustrate the practical aspects of monitoring through three instructive case studies, the Cardiac Arrhythmia Suppression Trial (CAST), the Antihypertensive and Lipid Lowering treatment to Prevent Heart Attack Trial (ALLHAT), and the Pamoja Tulinde Maisha (PALM) trial. Next, we summarize the difficulty of inference following a monitored trial. This includes the different ways to order the sample size, leading to different p-values and confidence intervals. We close the chapter with a brief summary of Bayesian monitoring methods, including predictive power.
