The World Health Organization defines pharmacovigilance (PV) as "the science and activities relating to the detection, assessment, understanding, and prevention of adverse events or any drug-related problem." Government agencies, clinical institutions, the pharmaceutical industry, and other entities manage strict pharmacovigilance programs as vital safeguards for supporting the early detection of safety signals. Current PV practices rely primarily on expert judgment and global introspection implemented in concrete business practices specific to an organization. Computational methods and decision-support systems can help standardize these business practices by specifying the evaluation steps human reviewers follow to discover unknown adverse reactions that remained unrecognized or were not fully understood in a medical product's pre-market period. Reviewers are frequently assisted by tools and methods that automate manual steps and allow more time for the constructive review. This special topic issue is intended to describe computational and systematic approaches to make this process more precise.
A complete decision-support system must address multiple data, user-related, and other challenges. To this date, multiple efforts have delivered solutions addressing more narrowly defined tasks rather than entire workflows. For example, a few methods have been proposed to encode drugs and adverse events, prioritize case reports for clinical and regulatory review, and detect duplicates in spontaneous reporting systems. These significant challenges and tasks in pharmacovigilance are part of more extensive multi-step processes that remain untapped from a (semi-)automated decision-support perspective. To collect the requirements and design a complete decision-support system incorporating Artificial Intelligence (AI) and other sophisticated algorithms, preliminary analyses by knowledgeable interdisciplinary teams are often necessary. As with any software development, end-users active participation in the development and evaluation phases is paramount. In most cases, this is the recipe for success as long as the right expectations are set from the beginning and the proposed solutions do not bring dramatic changes to existing workflows without proven benefit. Including a "human-in-the-loop" further ensures that a system's limitations, such as its ability to classify cases for a given task or minimize duplicates accurately, are carefully examined, and a quality assurance process to control them is in place. Ultimately, as no automated solution can be perfect, specific correction strategies and considerations are required if a new workflow incorporating a decision-support system mandates separate roles for humans and machines. Assessments of algorithm performance (e.g., validity, generalizability, absence of bias, and robustness in real-world settings with changing inputs), documentation, transparency, explainability, quality control with real-world data collection and monitoring, and algorithm change control are generally considered necessary for AI systems. These approaches are also needed when considering the maturity of decision-support systems.
We invite papers on the above topics and particularly encourage submissions that describe validated methods and active systems that have already demonstrated value and made significant contributions to pharmacovigilance by improving efficiency in established business practices, minimizing manual effort, maximizing discovery of safety issues, and standardizing existing processes transparently. Research and applications, brief communications, and review articles are the main types of submissions for this Research Topic. We also welcome case studies of system implementation and perspectives that may significantly contribute to the domain.
The World Health Organization defines pharmacovigilance (PV) as "the science and activities relating to the detection, assessment, understanding, and prevention of adverse events or any drug-related problem." Government agencies, clinical institutions, the pharmaceutical industry, and other entities manage strict pharmacovigilance programs as vital safeguards for supporting the early detection of safety signals. Current PV practices rely primarily on expert judgment and global introspection implemented in concrete business practices specific to an organization. Computational methods and decision-support systems can help standardize these business practices by specifying the evaluation steps human reviewers follow to discover unknown adverse reactions that remained unrecognized or were not fully understood in a medical product's pre-market period. Reviewers are frequently assisted by tools and methods that automate manual steps and allow more time for the constructive review. This special topic issue is intended to describe computational and systematic approaches to make this process more precise.
A complete decision-support system must address multiple data, user-related, and other challenges. To this date, multiple efforts have delivered solutions addressing more narrowly defined tasks rather than entire workflows. For example, a few methods have been proposed to encode drugs and adverse events, prioritize case reports for clinical and regulatory review, and detect duplicates in spontaneous reporting systems. These significant challenges and tasks in pharmacovigilance are part of more extensive multi-step processes that remain untapped from a (semi-)automated decision-support perspective. To collect the requirements and design a complete decision-support system incorporating Artificial Intelligence (AI) and other sophisticated algorithms, preliminary analyses by knowledgeable interdisciplinary teams are often necessary. As with any software development, end-users active participation in the development and evaluation phases is paramount. In most cases, this is the recipe for success as long as the right expectations are set from the beginning and the proposed solutions do not bring dramatic changes to existing workflows without proven benefit. Including a "human-in-the-loop" further ensures that a system's limitations, such as its ability to classify cases for a given task or minimize duplicates accurately, are carefully examined, and a quality assurance process to control them is in place. Ultimately, as no automated solution can be perfect, specific correction strategies and considerations are required if a new workflow incorporating a decision-support system mandates separate roles for humans and machines. Assessments of algorithm performance (e.g., validity, generalizability, absence of bias, and robustness in real-world settings with changing inputs), documentation, transparency, explainability, quality control with real-world data collection and monitoring, and algorithm change control are generally considered necessary for AI systems. These approaches are also needed when considering the maturity of decision-support systems.
We invite papers on the above topics and particularly encourage submissions that describe validated methods and active systems that have already demonstrated value and made significant contributions to pharmacovigilance by improving efficiency in established business practices, minimizing manual effort, maximizing discovery of safety issues, and standardizing existing processes transparently. Research and applications, brief communications, and review articles are the main types of submissions for this Research Topic. We also welcome case studies of system implementation and perspectives that may significantly contribute to the domain.