Aplicações da inteligência artificial na farmacovigilância : uma revisão de escopo

Abstract

Introduction: Pharmacovigilance is a strategic area of public health, responsible for identifying, analyzing, and reducing risks associated with the use of medicines. With the advancement of digital systems and the increasing volume of health data, the incorporation of Artificial Intelligence (AI) has emerged as a promising alternative to optimize processes, expand analytical capacity, and strengthen medication safety. In this context, there is a growing need to deepen the understanding of how these technologies are integrated into pharmacovigilance practices. Objective: To map the available scientific evidence on the applications of artificial intelligence in pharmacovigilance activities, identifying the techniques used, purposes, benefits, barriers, and gaps in the literature. Methodology: This is a scoping review conducted in accordance with the Joanna Briggs Institute (JBI) and PRISMA-ScR guidelines. The search was carried out in the PubMed, BVS, SciELO, Web of Science, and Scopus databases, using descriptors related to pharmacovigilance and artificial intelligence, combined with Boolean operators. Studies were selected according to predefined eligibility criteria, with screening performed by independent reviewers. Data were extracted and organized into standardized spreadsheets for descriptive analysis. Results: A total of 3,649 studies were identified, of which 28 were included in the final sample after the screening and eligibility stages. Most publications were concentrated between 2021 and 2025, with a predominance of studies conducted in countries with high technological development. The main AI applications included the detection and prediction of adverse drug reactions, the automated extraction of information from clinical texts, and the automation of processes related to case triage and coding. Among the techniques employed, machine learning, deep learning, and natural language processing were highlighted. The studies reported benefits associated with increased efficiency and improved capacity for detecting safety signals, including reductions of up to 26% in the need for manual signal validation and models achieving AUC values above 0.90 in the automated detection of adverse drug reactions, although limitations related to data quality, model interpretability, and ethical and regulatory challenges were also identified. Conclusion: The results demonstrate that artificial intelligence is being progressively incorporated into pharmacovigilance practices, providing clear benefits in the processing of large volumes of data and in the early identification of risks associated with medication use. Despite the identified potential, the adoption of these technologies is still occurring gradually, requiring greater validation in real-world settings and the strengthening of regulatory frameworks. In this context, artificial intelligence is expected to assume an increasingly strategic role in ensuring medication safety.