Semantic annotations are often used to enrich documents as clinical trials and electronic health records. However, the usability of these annotations tends to decrease over time due to the evolution of the domain ontologies. The maintenance of these annotations is critical for tools that exploit them (e.g., search engines and decision support systems) in order to assure an acceptable level of performance. Despite the recent advances in ontology evolution systems, the maintenance of semantic annotations remains an open problem. In this paper, we introduce, based on previous experiments, the main components of a multi-level approach towards the automatic maintenance of semantic annotations. We further provide examples for strengthening our propos