Sample Size Estimation and Power Analysis: Longitudinal Data
Controlled clinical trials often randomly assign individuals to treatment arms and follow them to collect measurements at intervals across a treatment period of fixed duration. This type of data are called longitudinal data. In such studies, one of popular primary objectives is to compare the change rate of the measurements over time between treatment arms. The repeated measurements over time span tend to be correlated within each individual and the analysis method to fit the time trajectory should appropriately incorporate the dependency. However, it is difficult or almost impossible to identify the true dependency structure among repeated measurements. Generalized estimation equation (GEE) method has been has been one of the most popular methods to fit the time trajectory of longitudinal data because it does not require specification of the true dependency structure. In this chapter, we discuss power analysis and sample size estimation for randomized controlled trials (RCTs) with longitudinal data using the GEE method to test on the difference in change rate between treatment. The sample size formula is easily converted to a power function. We consider the cases where the repeated measurements are continuous or dichotomous variable. These methods can be used for designing observational studies as well as RCTs.
