Menu Back to Poster-Presentations-Details

P212: AI Under the Hood: What Natural Language Processing (NLP) Means for Pharma -Moderna, Coherus Medical Affairs Use Case





Poster Presenter

      Walter Bender

      • Chief Technology Officer
      • Sorcero
        United States

Objectives

Assess the viability and maturity of AI Natural Language Processing for highly regulated pharmaceutical operations, specifically literature reviews in various drug development functional areas, including medical affairs and regulatory affairs.

Method

Comparison of man-hours, sensitivity, specificity, versatility/utility of outputs, and depth of insights for traditional manual literature search against Google AI developed NLP BERT data standard, subsequent BioBERT methodologies, as well as situational data (therapeutic area, indication, etc.).

Results

? The Intelligent Literature Monitoring (ILM) approach with integrated AI was the most favorable option for search strings with high volumes of search results (e.g. ”oncology” or “COVID”) ? Compared with manually searching and reviewing the literature, this method resulted in a time reduction of 88-92%, along with 99.8% sensitivity and 95% specificity ? Language Intelligence (LI) and continuous learning delivered double-digit absolute performance improvements across all study types, exceeding the 95% NPV threshold commonly accepted as that required for a regulatory-grade literature review solution ? For low-volume search results, (e.g. rare diseases) AI-driven NLP improved accessibility, distribution, communication, archiving, and sorting/filtering

Conclusion

Similar to industry transformations away from manual efforts, the use of ILM is a change that requires letting go of a traditional method and embracing new ones. As a traditional process, manual search and standard literature filtering/cognitive reviews vs. NLP is similar to clinical operations’ transformation from 100% source data review to risk-based monitoring, or from clinical pharmacology to pharmacometrics. The benefits of intelligent, well-executed NLP in life sciences include: ? Near-perfect command of all available literature, regardless of (overtime, exponential) volume ? Immediate relevancy and reduction of cognitive review of abstracts and content, in order to gauge contextual relevancy ? Machine learning -- The ability to refine the model, accuracy, and relevance over time, with system usage ? Freeing up human cognition for other, more strategic drug development operations