Medical recommendation may have the problems of multi-source heterogeneous data and multi-criteria of recommendation items in clinical practice. Considering the characteristics of medical recommendation, this paper defines the distance measure of different types of data in heterogeneous information systems and the effective processing of multi-source heterogeneous data is realized. Firstly, meanwhile the binary relationship in heterogeneous information systems is obtained according to the hybrid distance between two objects and the heterogeneous information rough set model is constructed subsequently. Then the multi-criteria recommendation is combined with the multi-criteria decision-making method(MCDM). And the multi-criteria rating of each item is aggregated by grey relational analysis(GRA) to transform multi-criteria recommendation into single-rating recommendation. Finally, three-way decision-making is presented on the basis of the heterogeneous information rough set model, and three-way recommendation is achieved based on the collaborative filtering method, which considers the cost of decision-making in the process of recommendation. Clinical practice data in the part of medical application is used to verify that the proposed model can provide a knowledge support for clinical diagnosis, which effectively reduces the cost of recommendation and improves the accuracy of recommendation.