Study protocol - NLP-DeVal: Development and validation of a natural language processing tool to enable clinical research in emergency and acute care medicine: retrospective cohort study

Within the eCREAM project, it has been proposed a study to develop and validate a language model (called eCREAM_LM) for the six languages of the project that is able to interpret EHR contents and extract relevant information for research purposes.

This study involves 35 centers from 7 different countries (Italy, Poland, Greece, Slovakia, Slovenia UK, and Switzerland).
The creation of the eCREAM_LM language model will entail training and finetuning different language models through exposure to both a massive amount (billions) of medical texts derived from the literature and a large quantity of non-annotated (millions) and annotated (5,000) clinical notes derived from EHRs.
The validation of eCREAM_LM will be done by assessing its concordance with a panel of experts in extracting crucial information from 3,000 clinical notes.
The study will last 24 months and will involve retrospective data collection.