Impact of Artificial Intelligence and Machine Learning on Adverse Event detection from real time sources

Abstract

The field of pharmacovigilance, which ensures drug and patient safety, faces numerous challenges in detecting and managing adverse events (AEs) associated with pharmaceutical products. Traditional methods of AE detection often rely on manual processes, which are prone to human error and resource intensive. In response to these challenges, there is a growing recognition of the potential of cognitive tools like artificial intelligence (AI) and machine learning (ML) techniques to augment the effectiveness and efficiency of AE detection processes. Firstly, AI and ML offer capabilities that can process vast amounts of data from diverse sources, including electronic health records, social media, and biomedical literature, to identify potential AEs more rapidly and accurately than manual methods. Furthermore, AI and ML hold promise in enhancing the efficiency of AE detection processes by automating repetitive tasks, triaging incoming reports, and prioritizing cases for further investigations. This automation not only accelerates the identification of high-priority AEs but also frees up human resources to focus on more complex tasks. Moreover, AI and ML algorithms can detect patterns and associations that may skip human observers, thereby enhancing the sensitivity of AE detection. By leveraging techniques such as natural language processing (NLP) and predictive modelling, these technologies can uncover hidden correlations. However, with the rise of social media and real-time chatbots, there is a great surge in real-time data generation. To be at pace with the information and identify risks for both patient and drug safety, it is of paramount importance to ensure the AEs are identified in real-time as well. Thus, it becomes more prominent to evaluate the available cognitive tools like AI and ML can also be credible enough to enhance effectiveness and efficiency in the AE detection process. This thesis comprises three research problems. First, it investigates the capability of AI and ML-based algorithms to enhance the accuracy of AE detection in real-time and from real-time data sources. Additionally, it also studies the moderating effect of the variable's languages, regions and disease areas on AI and its capability to impact the accuracy of AE detection. Second, it investigates the potential of the AI and ML-based algorithm in increasing the efficiency of the AE detection process in the overall individual case safety report (ICSR) process of pharmacovigilance, for cases arising from real-time data sources. Additionally, in this study too we explore the impact of regions, languages, and disease areas on the AI capability to enhance the efficiency of AE detection. Finally, it explores the possibility of using a domain-specific rule-based algorithm alongside AI models to enhance the selectivity and sensitivity of the AE detection. Additionally, our study compares and evaluates the selectivity and sensitivity between AI models and AI models in conjunction with domain-specific rule- based algorithms.