We design and develop bespoke solutions using state-of-the-art and proprietary NLP algorithms to automate the extraction of insights from the ever-growing medical literature. This allows to speed up systematic literature review processes with regulatory-grade standards.
Use case examples include:
- Automated systematic reviews on drugs efficacy and safety, on predictors and on methodologies
- Identification of KOLs and specialty centers in each geography
- Literature screening to map mechanisms of action and metabolic/disease pathways to generate hypotheses for drug repurposing
- Literature scanning and monitoring of case reports or rare adverse events
Medical literature mining for safety monitoring
Context: resource-heavy signal identification
Quinten Health is developing a literature-based safety monitoring system for a national regulatory agency in Europe.
Pre- and post-marketing surveillance need real-time robust signal identification to monitor the safety of drugs under review or on the market, based on submitted pharmacovigilance or trial data but also based on medical literature. This translates in up to 1,000 serious events to routinely track in millions of scientific publications, with new papers published every day.
A combination of limited human resources and large volumes of scientific publications to review makes it almost impossible to perform this task manually.
Quinten Health’s solution: NLP-based approach
To address this challenge, Quinten Health developed a bespoke AI solution leveraging Natural Language Processing (NLP) technologies to monitor, on a weekly basis, safety signals. This NLP solution can detect severe adverse events associated with any given drugs for any indication of interest. The results are displayed in the form of a dashboard for the medical reviewer to assess.
Impact: improved accuracy and efficiency
The end-to-end NLP-based solution for medical literature monitoring showed to be more comprehensive than manual review (up to +240% for some drugs/events combinations), and up to 49 times more accurate and 13 times less time-consuming than manual monitoring.