Author:

Abstract:

In longitudinal experiments a sample of patients is followed over some period of time, during which one or multiple responses are measured at several occasions. Such studies allow to study within-subject evolutions and to explore to what extent these evolutions depend on patient characteristics. The statistical analysis of longitudinal data requires specific tools and models which account for the association between measurements taken repeatedly with subjects. Such experiments also yield data structures far more complicated than in classical cross-sectional studies, as the same number of measurements is not always available for all subjects, nor are the measurements for all subjects taken at identical time points. Often, such unbalanced structures are the result of a dropout, i.e., patients leaving the study prematurely for various reasons. This does not only affect the analyses technically; it also seriously complicates the clinical interpretation of the obtained results. In this paper, the strengths of longitudinal studies are illustrated based on a series of examples, including experimental as well as observational studies. Emphasis is placed on the different research hypotheses that can be handled, on the advantages and disadvantages of some statistical analysis tools frequently applied in the medical literature, as well as on the interpretation of the final results.