In a research synthesis on the prognosis of depression, the outcome of interest has a multinomial distribution with the following typical categories: well, still depressed, dead, others. To analyse such data, we adapted the linear mixed effects model for multiple outcomes proposed by Berkey et al. (Statistics in Medicine 17, 2537-2550, 1998) in the meta-analysis context of a multinomial outcome. The fixed and mixed effect models, the methodology for the estimation of the parameters and results of an example on the prognosis of depression in the elderly are presented.