A Likelihood-Based Approach for the Analysis of Longitudinal Clinical Trials with Return-to-Baseline Imputation