A Linear Mixed Model for the Longitudinal Analysis of Difference Scores (303217)

Keywords: longitudinal analysis, difference score, linear mixed model

Background: When longitudinal data has little missing baseline data, analysis of difference scores is one method of normalizing the error terms, even if the original outcome is non-normal. Adjusting for the baseline value as a covariate enables estimation of difference scores, with adjustment for the starting value.

Objective and Methods: Derive the linear mixed model for difference scores, which will include terms for time, treatment group, interaction between time and treatment, and interaction between treatment group and baseline value.

Demonstrate how to use common statistical software, such as SAS, to prepare a data set for longitudinal analysis of difference scores. Present a SAS macro that uses Proc Mixed for analysis of difference scores, with adjustment for the baseline values of treatment groups. Further, explain how the between-group constrasts can be adjusted for multiple comparisons.

Results. Examples will be presented that show the trajectory of an outcome over time between treatment groups, in table and graphic format.

Conclusion. Outcome analysis, based on a LMM on difference scores with baseline adjustment, is an effective longitudinal analysis technique.