Abstract
<title>Abstract</title> <p>Background Herbal medicines are used together with prescription drugs by many patients across multiple clinical settings, yet the global pharmacovigilance reporting databases, the FDA Adverse Event Reporting System (FAERS), WHO VigiBase, EudraVigilance, and India's Pharmacovigilance Programme (PvPI), contain the herbal product individual case safety reports (ICSRs) characterized by significant botanical identity variability or inconsistency, co-medication disclosure gaps, and severe underreporting of the data. The machine learning and natural language processing methods have transformed signal detection for conventional pharmaceutical products but haven’t been applied to herbal product surveillance in any systematic way. Objective This review maps published artificial intelligence and machine learning methods applied to both herbal and botanical product pharmacovigilance signal detection, characterises the structural barriers in data quality that constrain their application, and also introduces the Minimum Information for AI-Assisted Herbal Pharmacovigilance (MIAHPV), a newly proposed 7-point framework to evaluate how ready herbal ICSRs are for AI-based analysis. Methods The review followed PRISMA ScR guidelines and was based on searches in PubMed/MEDLINE and the Cochrane Library, covering studies from January 2010 to March 2026, supplemented by a grey literature search; database access was limited to freely available sources as the author is an independent undergraduate researcher without institutional subscription access to Embase, Scopus, or Web of Science. We focused on studies where artificial intelligence or machine learning had been used with pharmacovigilance data on herbal or botanical products. We also included papers that discussed data quality problems, especially where these issues could make it harder to detect signals in herbal ICSRs. A structured approach was then used to develop the MIAHPV framework criteria, based on the needs of AI methods and the common data quality gaps identified in herbal ICSRs across the selected studies. Results Across the studies we reviewed, including seven studies identified through our own database search that applied AI or NLP methods to herbal or dietary supplement safety data, we did not find any that applied AI-based signal detection specifically to herbal product ICSRs as the main exposure, which points to a clear gap in the current methods. Three common data quality issues came up repeatedly: inconsistent reporting of botanical identity, incomplete information on co-medications, and general underreporting. When comparing different regions, we also noticed that regulatory approaches to coding herbal ICSRs vary across agencies such as the FDA, EMA, WHO UMC, and CDSCO. Based on these findings, the MIAHPV framework outlines seven practical criteria. These include completeness of botanical identity, disclosure of co-medications, use of MedDRA-coded outcomes, reporter qualification, product authentication, plausibility of time to onset, and an initial causality assessment. Using these criteria, herbal ICSRs can be grouped into three levels based on how suitable they are for AI-based analysis. Conclusions AI-based approaches to pharmacovigilance in herbal medicine appear promising, but their use is currently limited by widespread data quality issues in existing reports. The MIAHPV framework offers a starting point by defining a minimum set of information that can support both improving existing data and guiding future reporting systems. The next step will be to test this framework using real-world ICSR datasets at the national level.</p>