Artificial Intelligence-Enhanced Pharmacovigilance for Diethylene Glycol Contamination in Paediatric Medicines: A Systematic Narrative Review and Proposed Computational Framework for Real-Time Outbreak Detection and Supply Chain Surveillance

Diethylene glycol (DEG) contamination of paediatric oral liquid medicines constitutes a recurrent, preventable global public health crisis. Since 2022 alone, WHO-confirmed outbreaks across The Gambia, Uzbekistan, Indonesia, and India have claimed the lives of more than 300 children. Despite over eight decades of documented incidents— beginning with the 1937 Elixir Sulfanilamide disaster—conventional pharmacovigilance systems have repeatedly failed to detect, contain, or prevent these outbreaks with sufficient rapidity to avert mass mortality. Artificial intelligence (AI) and machine learning (ML) offer transformative potential for real-time signal detection, supply chain anomaly identification, and predictive risk stratification in pharmaceutical quality surveillance.Objectives: To: (i) synthesise the global epidemiological, clinical, and toxicological evidence on DEG poisoning across twelve major outbreak events spanning 1937–2025; (ii) characterise the systemic pharmacovigilance and regulatory failures that have enabled repeated tragedies; and (iii) propose a validated, five-component AI-driven computational framework specifically architected to address identified failure domains.Methods: We conducted a pre-specified systematic narrative review of peer-reviewed literature, WHO Medical Product Alerts, regulatory communications, epidemiological field reports, and grey literature pertaining to DEG and ethylene glycol (EG) pharmaceutical contamination from 1937 to June 2025. Search strategies were applied across PubMed/MEDLINE, Scopus, WHO IRIS, Cochrane Library, and Google Scholar using MeSH-mapped terms. Supplementary AI/ML pharmacovigilance literature was systematically searched to inform framework design. Data were extracted across standardised domains: outbreak epidemiology, pathophysiological mechanisms, clinical presentation, management outcomes, regulatory responses, and identified failure points.Results: Twelve major DEG outbreak events were identified across ten countries, accounting for over 800 documented deaths, with children under five years disproportionately represented. Consistent root causes across all outbreaks encompassed six recurring failure domains: excipient adulteration, over-reliance on unverified Certificates of Analysis (CoAs), absence of mandatory in-house testing, inadequate regulatory inspections, opaque supply chains, and delayed alert and recall systems. Clinically, DEG produces a characteristic biphasic illness progressing to high-anion-gap metabolic acidosis, proximal tubular necrosis, and multiorgan failure; mortality is strongly time-dependent. Early fomepizole and haemodialysis significantly reduce case fatality rates. No outbreak reviewed employed AI-assisted surveillance or supply chain monitoring. The proposed AI framework comprises: (1) NLP-based real-time pharmacovigilance signal mining; (2) Graph Neural Network supply chain traceability; (3) CNN-assisted portable spectroscopic quality screening; (4) Federated Learning for cross-jurisdictional surveillance; and (5) LLM-powered regulatory alert synthesis.Conclusions: DEG poisoning remains a race against death in which current pharmacovigilance systems consistently fail to intervene in time. The proposed AI-driven framework, if deployed within an international regulatory architecture, offers a credible, scalable, and privacy preserving pathway to preventing future paediatric fatalities from preventable pharmaceutical contamination. Immediate investment in digital pharmacovigilance infrastructure—particularly in low- and middle-income countries—is a global health imperative.