This research work aims to improve the sanitary transportation in the Rhône-Alpes region in France. The transportation of disabled people from their homes to medical institutes corresponds to a classic operational research problem : the dial-a-ride problem (DARP). In the previous work on the subject, the framework developed in [3] has allowed to solve real-life problems up to 600 users. The goal of this work is to reduce the computation times and increase the size of the problems by taking advantage of the specificities of the problem. Recent works done in [1], proposed a novel approach to solve routing problems. Usually, the research is about designing strong heuristics tuned well enough to give satisfactory results across a large variety of instances. But in their work, most of the efforts were spent on understanding the structural characteristics of the problem prior to building the heuristic. By knowing the structural differences between good and bad solutions, it is possible make efficient heuristics which integrate this knowledge in their design. This knowledge-guided method allows to solve very large instances of vehicle routing problem (VRP), up to 30000 users within reasonable computation times. The goal of this study is to improve the performance attained in [3] on real-life problems by extending the work done in [1] from the VRP to the DARP.