Keywords: Natural Language Processing, Network Analysis, Exploratory Analysis
Post-market surveillance primarily focuses on the statistical evaluation of safety patterns towards the verification of actual signals. This usually includes the calculation of disproportionality analysis metrics that indicate whether an adverse event is more commonly reported for one product than others. Although such analyses have efficiently supported the identification of real safety concerns in the past, they are limited by the implementation of arbitrary thresholds for signaling, the lack of multi-dimensional exploration, and the sole use of coded data. The review of large datasets conducted by medical epidemiologists (following either a disproportionality value exceeding a certain threshold or a need for performing particular safety tasks) requires additional investigation of various sources, information retrieval from free-text narratives, and data collection for multiple parameters. Therefore, medical review requires the development of different approaches and the use of techniques ranging from advanced text analytics to compelling data visualizations. We present the use of particular natural language processing (NLP) tools that mine the free-text narratives of post-market reports; extract diagnostic, medication, and time information; and generate meaningful visualizations. We also demonstrate the utilization of the NLP output in a network analysis tool for exploratory and confirmatory analysis. Particular examples from the safety surveillance domain are discussed.