Multilevel and longitudinal data analysis

The course introduces the analysis of multilevel and longitudinal data (clustered data) for continuous and binary outcome measures. Advantages and challenges with the dependencies in clustered data will be discussed. Longitudinal study designs, in which individual measurements are taken repeatedly over time, will be given special focus.

The course covers descriptive statistics for multilevel and longitudinal data, covariance and correlation, variance components, and statistical analyses by linear mixed models. For longitudinal data, analysis of summary measures (e.g. AUC, linear slope, min/max values) and models with a categorical as well as a continuous time variable will be presented.

In addition, marginal models where the correlation structure for the repeated measurements is specified directly and not by the means of a multilevel model, including generalized estimating equations (GEE), will be discussed. How to choose the model that gives the best fit to the data, will be emphasized. Methods for longitudinal data from randomized studies with a pre-post design will be discussed particularly. The course includes exercises using the Stata software.

Knowledge

After successful completion of this course the student should

• be aware of the nature of dependency in clustered measurements, and how this dependency influences the approach to analysis and modeling, as well as the consequences of ignoring this dependency
• have knowledge of marginal and multilevel models (linear and logistic mixed models) for clustered data, and the difference between these models
• have knowledge of how and when these methods can be applied in medical research projects

Skills

After successful completion of this course the student should be able to

• summarize multilevel and longitudinal data by simple descriptive analyses and graphical displays
• identify clusters and potential dependency by inspecting the design and/or the resulting data set for a study
• identify the appropriate statistical method for analyzing a set of clustered data
• identify an appropriate correlation structure for the time dependency in longitudinal data
• independently perform a statistical analysis for multilevel and longitudinal data by the means of statistical software (Stata)
• evaluate the assumptions made on the applied model or method
• interpret and critically evaluate the results from the statistical analysis
• present the results in a format applicable for publication in a scientific medical journal

General competence

After successful completion of this course the student should

• be able to evaluate application of statistical methods for analyses of multilevel and longitudinal data in medical research projects

Lectures and exercises using Stata.