Updated: Oct 31
Challenges in Pharmacovigilance Literature Monitoring
With the large and growing volume of scientific literature available, automating literature monitoring activities is becoming a necessity for companies large and small. Despite the clear value the literature provides in the detection of novel safety events and safety signals, the task remains laborious and the yield is low.
As an example, the European Medicines Agency (EMA) performs medical literature monitoring activities of a groups of medications, and has issued interesting results in the recently published EMA Report on Pharmacovigilance Tasks: in 2019 the agency’s medical literature monitoring effort yielded 1.8% unique cases from screening just over 355K literature articles. As volumes grew in the years since, by 2022 the effort yielded 0.85% unique cases, with approximately twice the screening volume as 2019.
Keeping pace with the increase in literature volumes is crucial to ensuring the quality of the safety monitoring process so that unique adverse events can be detected. It is no surprise then that reliable automation for pharmacovigilance activities, in particular applications of artificial intelligence, continues to gain traction and remain an active topic of engagement in industry groups and academia.
How Automation can Help Medical Literature Monitoring
To better answer this question, lets take a look at each stage of the medical literature monitoring workflow:
Database Search and Deduplication of Results
To ensure the quality of results it is important that searches cover multiple relevant scientific databases. For example, guidelines from EMA’s GVP Module VI state:
“It is best practice to have selected one or more databases appropriate to a specific product.”
A single consistent query can save time in developing and maintaining query strings that are database-specific, and also give the ability to run a single search “once” against many databases.
Duplicate detection in literature monitoring is a critical aspect of ensuring data integrity and reliability. Identifying and eliminating redundant information prevents skewed analyses and ensures the accuracy of adverse event reporting.
Duplicate detection algorithms employ advanced techniques such as text similarity measures and rules-based methods to identify instances where the same information is reported across multiple sources. These algorithms must account for variations in language and formatting to effectively identify duplicates.
Ranking and Filtering Results for Literature Screening
After results are retrieved, they can be ranked or filtered according to the requirements of the workflow (ICSR, aggregate reporting, etc).
AI automation can assist with predictions that replicate the workflows of pharmacovigilance professionals. Biologit MLM-AI implements predictions that correspond to what drug safety and signal detection specialists are looking for in their workflows.
Users can rank or filter according to these predictions to reduce the volume of results to be screened or to perform more relevant and focused quality control processes.
Safety screening assessments
Automation can also help improve the quality and speed of assessments for safety events. Natural language processing techniques can identify and extract key components of the article, and machine learning methods can help extract and highlight, or summarize key passages for more focused assessments.
Extracted entities can then be used to pre-populate details for case entry, or for the generation of aggregate reporting.
Automation and AI in Medical Literature Systems: How to Choose?
There is significant complexity in the implementation of AI systems from scratch: sourcing and labeling data, design, implementation, validation and monitoring of a fit for purpose solution requires a diverse skill set and extensive domain knowledge. Partnering with a vendor for medical literature automation can help deliver a faster return on investment and a simpler implementation.
What should companies consider when choosing a partner for their project? Below are some key points:
Regulatory guidance for AI in pharmacovigilance
Emerging and established regulatory guidance and industry best practices need to be constantly monitored and incorporated into the AI development lifecycle of the solution to ensure audit readiness and compliance.
Implement robust AI transparency and governance
For AI systems to stay healthy, regular monitoring of its performance and constant attention to its quality is necessary. This is reflected in regulatory guidance and platforms should provide a high level of auditability on their governance processes. Ideally these should be embedded in industry standard quality frameworks that cater for non conformances, risk and change management, and validation.
Further, an approach that ensures transparency on the development process and operations of the automated literature monitoring platform help gain user trust and reduces risk by providing visibility on key project decisions.
📖Learn more: At Biologit we developed and made publicly available artifacts explaining the objectives of AI technology available in Biologit MLM-AI, its intended uses and limitations, for example:
For users: product guides and training.
For technical teams and auditors:
Model fact sheets summarizing model specification and intended use.
A Technical paper describing our approach with experimental results.
Compliance and validation
Computer systems validation (CSV) and compliance to industry standards are non-negotiable requirements for any pharmacovigilance solution, and no different for intelligence automation features. In particular platforms should demonstrate emphasis of validation activities in automated features, and how the platform remains compliant with standards such as CFR-11 and GVP.
Oversight by design (Human in the loop)
Ensuring automated actions - whether rules based or machine learning - have the necessary human oversight is necessary for ensuring the system continues to meet specifications over time and can be validated. To that end, the literature automation solution must ensure adequate visibility and traceability of automated decisions allowing for whatever levels of oversight are necessary for the project.
Intelligent Literature Monitoring with Biologit MLM-AI
biologit MLM-AI is a complete literature monitoring solution built for pharmacovigilance teams. Its flexible workflow, unrivalled scientific database, and unique AI productivity features deliver fast, inexpensive, and fully traceable results for any screening needs.