Back to Agenda
Session 7: Unexpected Issues in Pharmacoepidemiology Studies Applying Natural Language Processing to Clinical Notes
Session Chair(s)
Nirosha M. Lederer, PhD, MS
Head, US Government Partnerships; Senior Director, RWE Strategy
Aetion, United States
Key to driving data fitness for regulatory decision-making is improving the usability of unstructured notes, which are rich with clinical nuance but difficult to curate and analyze. Using the Sentinel Innovation Center's Multi-source Observational Safety study for Advanced Information Classification using natural language processing (MOSAIC-NLP) project as a case study - this session will describe strategies to address unexpected methodological challenges and discuss how to advance the use of NLP of clinical notes to support population-based pharmacoepidemiology studies. This project is applying NLP to unstructured EHR data from over 100 million patients across more than 100 health systems to identify outcomes, extract confounders, and contextualize longitudinal data.
Learning Objective : - Identify how to utilize clinical notes to create a representative NLP training sample
- Design the taxonomy for annotation of the training set
- Create transparency and reproducibility through appropriate documentation
- Assess model generalizability given health system differences
- Evaluate epidemiological issues when using NLP to analyze clinical notes
Speaker(s)
Speaker
Dena Jaffe, PhD
Oracle, Israel
Lead Data Strategist
Hasham Ul Haq
John Snow Labs, United States
Machine Learning Engineer
Speaker
Darren Toh, DrSc, FISPE
Harvard Medical School and Harvard Pilgrim Health Care Institute, United States
Professor
Have an account?