JIST

Toxicology in the Age of Artificial Intelligence

    Anand Mugadlimath       Mandar R Sane

Artificial intelligence (AI) is transforming toxicology from bedside decision-making to population-level toxicovigilance. By integrating machine learning (ML) with large clinical datasets, AI offers opportunities to improve risk assessment, prognostication and early detection of emerging toxic threats. However, implementation also raises ethical, legal and practical challenges, particularly in low- and middleincome countries such as India. Traditional toxicology relies on clinical expertise, pharmacological knowledge and epidemiological evidence.[1,2] ML now enables rapid analysis of poison-centre databases, improving prediction of poisoning severity and patient outcomes.[3,4] Studies involving methadone, lithium, calcium channel blockers and mixed poisonings demonstrate encouraging predictive performance.[3–5] AI should therefore complement—not replace—clinical judgement.

Poison information centres are ideal settings for AI-assisted decision support because clinicians frequently manage incomplete histories and diagnostic uncertainty. Models integrating demographic, clinical and laboratory variables can improve triage, estimate risk, recommend monitoring and support antidote or ICU decisions.[1,3,4] Continuous learning systems may further improve performance as poisoning patterns evolve.[3] AI also strengthens toxicovigilance by identifying changing poisoning trends, emerging toxic agents and vulnerable populations.[2] ML algorithms outperform conventional statistical approaches in identifying poisoning risk factors and may facilitate early outbreak detection, surveillance of counterfeit medicines, evaluation of regulatory i nterventions and targeted prevention programmes.[2,4]

Predictive models using routinely available clinical variables have shown promising performance in estimating severe outcomes and mortality.[3–5] Such tools may support decisions regarding intensive care, haemodialysis or continuous renal replacement therapy, optimize resource allocation and improve clinical audit and training.[3–5]

Despite these advances, important ethical and medico-legal issues remain. Algorithms developed using non-representative datasets may perform poorly in Indian populations, making local validation essential. AI systems should be transparent and explainable so clinicians understand factors influencing recommendations. Outputs should remain advisory, with final responsibility resting with qualified healthcare professionals.[3] Strong data governance and protection of sensitive poison-centre information are also essential.

India bears a substantial burden of poisoning due to pesticides, pharmaceuticals, household chemicals and envenomation, yet poison information systems remain fragmented.[2] A pragmatic roadmap includes creation of a standardized national poisoning database, development and validation of India-specific predictive models, integration of AI into routine clinical workflows, incorporation of AI literacy into medical education, establishment of regulatory frameworks for AI-based decision support and equitable deployment in resourcelimited settings using routinely availableclinical variables.[2–5]

Artificial intelligence offers genuine opportunities to improve poisoning care through better prognostication, smarter triage and enhanced toxicovigilance. Its successful implementation, however, requires rigorous validation, transparency, ethical safeguards and robust regulatory oversight.[2,3] The future of toxicology will depend not on replacing clinicians with AI, but on developing trustworthy human–AI partnerships that deliver safer, faster and more equitable poisoning care in India.

References 1. Burwinkel H, Keicher M, Bani Harouni D, et al. Decision support for intoxication prediction using graph convolutional networks. arXiv. 2020;2005.00840.[1]

2. Shirmardi K, Riahi A, Heshmat R, et al. Comparison of machine learning algorithms to predict intentional and unintentional poisoning risk factors. Heliyon. 2023;9(6): e17131.[2]

3. Karami Z, Sadeghi M, Farrokhian A, et al. Outcome prediction of methadone poisoning in the United States: implications of machine learning in the National Poison Data System (NPDS). Drug Chem Toxicol. 2024;47(5): 556 563.[3]

4. Zhang Y, Li X, Chen J, et al. Machine learning based prognostic prediction models in calcium channel blocker poisoning. Sci Rep. 2025;15:94395.[4]

5. Li J, Wang H, Xu Q, et al. Machine learning based outcome prediction for patients with acute poisoning receiving continuous renal replacement therapy. Longdom Proceedings. 2024;Sep:1 4.[5]