The goal of Clinical Decision Support (CDS) is to link relevant medical information to meet physicians' needs for taking better care of their patients. An effective CDS system takes medical records as input and returns the most relevant full-text biomedical articles to provide clinical decision supports for the physicians. Existing CDS systems mainly consist of three processing stages i.e. Query Reformulation, Retrieval and Reranking. Most of the existing CDS systems are only implemented based on classical keyword-indexing retrieval models such as BM25, PL2 and BB2, without taking semantics into account. To this end, this paper develops a novel CDS approach based on the semantic vector representations of both documents and queries. Experimental results on the standard Text REtrieval Conference (TREC) CDS track dataset show that the performance of a CDS system can be improved by integrating the semantic information.