Skip to main content

Case Study: How CSL Behring Uses Natural Language Processing To Improve MedDRA Coding

Situation: CSL Behring’s pharmacovigilance department processes AE verbatim reports. These are received in natural language for analysis and disambiguation, and are coded according to MedDRA standardized terminology. Autocoding only worked for 30% of the verbatims; 70% were manually coded. As CSL Behring works with rare diseases, more than 90% of the verbatim reports are unique and never repeated. CSL Behring wanted to see if natural language processing (NLP) could reduce this time-consuming manual effort and improve the quality of MedDRA coding.

Solution: CSL Behring worked with the IQVIA NLP team (Linguamatics) to develop a proof of concept. They extracted verbatim reports and codings from their AE database and created randomized training and test data sets. Linguamatics used the training data set to develop NLP queries using the IQVIA NLP platform. The queries were run against the test data set, and the coding efficiency and quality were compared with the manually coded data.

Success: This proof-of-concept experiment doubled the level of autocoding from 30% to 60%, with a very low level of mismatch. Examination of mismatches and non-coding showed that the NLP-proposed MedDRA codes were often of better quality and consistency than the previous manual codes. Building on this success, CSL Behring is planning the next version of the tool, which will be better integrated into the CSL Behring safety system, with ongoing machine learning improvements.

To access the case study, you will first need to fill out a short webform:

Ready to get started?

Request a Demo

Questions? Ask our experts