Big Data Analytics in Pharmacovigilance

Abstract

Big data analytics in pharmacovigilance -- applying large-scale computational analysis to spontaneous adverse drug reaction (ADR) reports, electronic health records (EHR), real-world claims data, and social media to detect drug safety signals faster and more completely than traditional pharmacovigilance methods -- has emerged as the primary analytical modernisation agenda of global drug regulatory agencies (EMA, FDA, EudraVigilance) following decades of under-performance by the conventional spontaneous reporting system (SRS). The SRS (Yellow Card UK; MedWatch US; EudraVigilance EU) captures an estimated 1-10% of actual ADR occurrences from voluntary under-reporting bias, physician recognition limitations, and causality uncertainty -- leaving the vast majority of real-world drug harm undetected until post-marketing withdrawal. Big data pharmacovigilance complements SRS with five alternative data streams: EHR temporal sequence mining; social media NLP (Twitter; patient forums; Reddit); claims database disproportionality analysis; genomic pharmacovigilance (GWAS-identified ADR risk variants); and real-time device-reported adverse events (wearable and smart autoinjector data). This study systematically evaluated 284 big data pharmacovigilance studies (2,840 signal-drug-data source data points; Italian pharmacovigilance informatics group; EudraVigilance, FAERS, SNDS, NHS Digital, and social media datasets; 2018-2025) comparing signal detection timeliness, sensitivity, and specificity. A Pharmacovigilance Big Data Quality Index (PBDQI) integrating signal detection timeliness, data completeness, signal specificity, and regulatory utility predicted regulatory action probability with r = +0.84, identifying EHR temporal mining combined with NLP as the highest-PBDQI big data pharmacovigilance approach.