Large sources of structured knowledge are available in many domains, enabling the construction of applications requiring relational knowledge. But in spite of the apparent availability of relational content, the semantics and granularity of these sources don't always match the requirements of specific tasks. Yet even when the coverage of explicit relational knowledge in a source seems inadequate, there may be implicit knowledge in the complete space of relation instances. In this paper, we show that explicit relation instances in the Unified Medical Language System (UMLS) are insufficient for our task of detecting relations between concepts in Electronic Medical Records. But by mining UMLS for relational paths between pairs of concepts in a training set, then using generalizations of those paths as features in a classifier, we achieve better results on the relation detection task than using state-of-the-art, corpus-based relation extractors. We further show that by combining the mined path-based features with the features used in the corpus-based extractors, we achieve even better performance.