Longitudinal data set commonly contains various extent of missing data. Two approaches, maximum likelihood based linear mixed-effect model (LMEM) and the sequential regression multivariate imputation (SRMI) were used to analyze a selected longitudinal data set with about 20% missing value at baseline measurement and close to 30% missing at five year follow up.

Data from a total of 373 patients was collected from in a prospective multicenter study of resective epilepsy surgery. In order to determine the relationship between changes in depression and seizure control status after resective surgery, data set was analyzed using 1) incomplete data set by SAS proc MIXED representing LMEM approach and 2) complete data set generated by multiple imputation using IVEware representing SRMI approach. Three major steps in fitting of LMEM 1) Selection of a preliminary mean structure 2) Selection a random-effects structure and correlations structure and 3) Assumption diagnostics were illustrated in details. For SRMI approach, ten complete data sets were generated by multiple imputations using software IVEware. Each complete data set was then analyzed using LMEM same as the first approach. SAS MIANALYZE procedure was subsequently used to combine the results from imputed multiple data sets to generate valid statistical inferences about the parameter.

The results from both approaches are similar and all indicated that the better seizure control (determined by length of seizure free period after surgery), the less likely have higher Beck Depression Index (BDI) score which represents moderate or severe depression symptoms.