From Data Silos to Standardized, Linked, and FAIR Data for Pharmacovigilance: Current Advances and Challenges with Observational Healthcare Data

Abstract

Pharmacovigilance (PV) encompasses all data gathering and processing activities related to the detection, assessment, understanding, and prevention of adverse effects throughout the entire life cycle of drugs [1]. The current era of “data explosion” or “big data” affects the entire spectrum of health sciences, including PV [2]. In particular, the data employed for PV have been recently extended, considering not only traditional/dominant data sources, i.e. spontaneous reporting systems, clinical trials, and the scientific literature, but also observational healthcare databases (i.e. electronic health records [EHRs] and administrative claims) with potential linkage to genetic data, as well as social media platforms and mobile health (mHealth) apps [3]. Notably, large-scale observational healthcare data networks are being increasingly established nationally [4, 5], and internationally (e.g. the European Health Data and Evidence Network, http://www.ehden .eu/). Providing access to huge observational healthcare data holds much promise, as it enables the answering of complex questions about care decisions and patient outcomes [6]. This advancement comes with the benefit of broadening the search space for real-world evidence in PV, but also brings new requirements and challenges, both scientific and technical, given that the above sources are not designed to serve PV aspects per se (i.e. cases of secondary data use). Εspecially in the scope of the concurrent use of multiple data sources for knowedge discovery, it is important to address the underlying data heterogeneity and follow the paradigm shift from data silos (i.e. isolated and disparate data systems) to uniformly represented and, thus, exploitable data structures [7, 8]. Data heterogeneity is an inherent characteristic of observational healthcare databases and EHRs in particular [9], resulting in questionable credibility and reproducibility of the results obtained through their analysis [10]. To study patient outcomes with EHR data, appropriate cohorts have to be defined using standard/reference codes (as part of the so-called semantic normalization process) for representing diagnoses, clinical procedures, drug prescriptions or dispensing events, and laboratory examinations. In the current issue of Drug Safety, Lee et al. [11] present the development of a controlled vocabulary-based dictionary, aiming to facilitate the conduction of multi-center EHR-based PV studies. In particular, the intended use of the proposed dictionary concerns adverse drug reaction (ADR) signal detection and evaluation. As the authors argue, “an ideal ADR signal-detection method should be able to investigate all drugs by all signs, all symptoms, and all abnormalities; therefore, systematic mappings between various EHR data resources and ADR-related standardized terms are essential for extensively detecting ADR signals from EHR data”. In this regard, efforts aiming to map or link medical terminologies/vocabularies for PV research are very important, especially when the mappings are made available for use by the wider research community, which is the case for Lee et al. [11]. In the scope of multi-center EHR-based PV studies, various data elements such as diagnoses, medical procedures, medications, and laboratory tests shall be expressed uniformly by selecting and combining codes from diverse vocabularies as well as proprietary coding schemes. Using reference terminologies for this mapping process facilitates standardization and semantic interoperability [12]. Given that each vocabulary reflects a different perspective for expressing and organizing the concepts of interest, limiting the number of vocabularies used limits the population This comment refers to the article available at https ://doi. org/10.1007/s4026 4-018-0767-7.