Application of Mixed-Effect Mixed Distribution Model Substance Abuse Studies

This study demonstrates application of the recently developed Mixed-Effect Mixed Distribution Model (MEMDM) to substance abuse studies. Common problems with continuous outcome measures in longitudinal studies include a spike at zero in the repeated measures, skewness of the non-zero values of the measures, and intra-class correlation or ICC. The MEMDM accommodates these problems (i.e., extra zeros, skewness, and ICC) simultaneously. The MEMDM consists of two sub-models: the logistic regression that models the probability of occurrence of non-zero outcome values; and the linear regression model that explains amount of the outcome as a function of linear predictors. Radom effects are incorporated in each of the two sub-models to account for unobserved heterogeneity, and the random effects in the two sub-models are allowed to be correlated. An example of application of the MEMDM is demonstrated using a sample of 430 crack cocaine users recruited from the Dayton, Ohio, area between July 1996 and August 1997. The outcome measure under study is the depression score measured by Beck Depression Inventory (BDI) over an 8-year observation period. SAS macro MIXCORR is applied for model estimation.