he main objective of longitudinal clinical trials is to compare the treat- ment effects depending on an outcome variable. Standard techniques can be used when all intended measurements for all subjects are available. Some patients may leave the study prematurely resulting in monotone missing data (dropout). A major problem arises when the probability of dropout is related to the outcome variable, this often reffered to as informative or non- ignorable dropout. Ignoring the missing data in this case leads to biased estimates of treatment effect differences.  This paper proposes and developes the stochastic EM algorithm to obtain valid estimates of treatment compar- isons. The proposed algorithm is a variant of imputation approaches, which are conceptually and practically simple and are commonly used by practi- tioners. Simulation studies are conducted to evaluate the proposed approach and to compare it with three common approaches; the "last value carrying forward" (LVCF) approach, the "all available data" (AAD) approach and the "partial imputation approach" (PI). Simulation results show that the stochastic EM approach provides unbiased treatment comparisons, or at least "less" biased, comparable to the other three approaches. The stochas- tic FM approach, also, can be used to estimate individual treatment effects, not only comparing treatment effects, as the case with other approaches.