Harnessing Transformers for Detecting Adverse Drug Reaction and Customized Causality Explanation using Generative AI

Pharmacovigilance plays an important role in monitoring safety of pharmaceutical products. There is an abundance of unstructured data available in social media and online reviews that can provide valuable insights into adverse drug reaction (ADR). This research paper aims to explore the efficient techniques of pharmacovigilance in identifying ADRs from unstructured data available in social media and reviews. This research work proposes the use of a transformer-based model for ADR classification, followed by named entity recognition (NER) to identify the causality between drugs and adverse effects. Furthermore, the output of NER is provided as a customized prompt to the generative AI model to generate a user interpretable explanation. Enhancing the accuracy of NER is carried out using the SciBERT model, which is specifically designed to capture drug and effect names. By utilizing this combination of techniques, the research aims to improve the efficiency and effectiveness of pharmacovigilance processes, enabling earlier detection and reporting of ADRs. To our knowledge, previous works have not explored the application of generative AI based customized explanation of causality. Patients, healthcare professionals, regulatory agencies, and pharmaceutical companies can benefit from the timely and accurate identification of ADRs, allowing for improved drug safety monitoring, better decision-making regarding drug usage, and the development of proactive measures to mitigate risks.