Longitudinal Data Analysis with Informative Time Measurements: a simulation study

Longitudinal data analysis plays a key role in a multiplicity of distinct areas, including medicine. One of the great difficulties in this type of study is related to different observation times for different individuals, times that are treated as independent of the response variable. An even greater difficulty occurs when the different observation times are related with the response variable. For example, the doctor decides to mark more, or fewer, appointments according to the patient's state of health. In cases where observation times and response variables are related, a simple longitudinal analysis will produce biased estimators and, consequently, uncertain conclusions. Therefore, it is necessary to develop new methodologies that allow the inclusion of this characteristic. We intend to present here some alternative models, which fit into the problematic, demonstrating their differences through a simulation study.