Statistical researches regarding health-related quality of life is a major challenge to better evaluate the impact of the treatments on their everyday life and to improve patients' care. In the literature, the mixed models based on the item response theory (IRT) are proposed to analyze directly HRQoL data from the questionnaires given to the patients. First, we use a recent classification of regression models for categorical data to discuss about a selection of IRT models for the longitudinal analysis of health-related quality of life in cancer clinical trials. Through methodological and practical arguments and an illustration on real data, the adjacent and cumulative models seem particularly suitable for this specific application. Then, a simulation study is carried out to compare the linear mixed model classically used to the most suitable proposed models. These simulations are a complement of other works concerning the comparison between classical test theory models and IRT models. This study is performed on the random part of mixed models and shows the IRT models are more precise. In opposite to the linear mixed model currently used, the IRT models are sensitive to the model used to generate the data. Between two IRT models used on the data, we recommend to consider that which does not take into account the random effect. If both considered the random effect, the choice of model depends to preference of user following the arguments of the first part of this work.