Pharmacovigilance, a set of activities related to monitoring the safety of medicinal products throughout their life-cycle, is crucial for public and personal health. Traditionally, post-marketing pharmacovigilance is done via spontaneous reporting: the collection and analysis of reports of suspected adverse drug reactions. We propose a novel approach to represent the two major types of concepts found in spontaneous reports, the drugs and reported reactions, using distributional semantics based on patterns of co-reporting. As proof-of-concept, we demonstrate the meaningfulness of the built representations using manually curated hierarchies of clinical relevance. This preliminary work may lay a foundation for novel down-stream improvements in the science of pharmacovigilance, as has been observed when distributional representations of words were introduced in the field of text analysis.