Online health community can break the limitations of time and space to provide medical and health information for users, meanwhile can promote the formation and evolution of friendships between users. This paper takes the largest diabetes community in China - Sweet Home as the research, and uses the dynamic social network analysis, which is based on stochastic actor-oriented models, to study the impact of individual attributes and network structure on the evolution of users' friendships. It is found that in sub-forums that highly relate to diabetes, basic user attributes such as gender, age, and disease type have significant impacts on the formation of relationships, while in sub-forum of sharing and accompanying, the detailed attributes such as the number of friends, online time, bonus points have a significant impact on the formation of relationships. However, transitive triads have no significant influences on the formation of friends.