PatientDatabase conversion for MOTOR model compatibility (FEMR 0.1.16 - jax) from FEMR 0.2.4 / meds 0.3.3 / meds-reader 0.1.16 / meds-etl 0.3.12
Unfortunately etl_generic_omop (via https://github.com/som-shahlab/femr/blob/0.1.16/src/femr/etl_pipelines/omop.py) was resulting in segmentation faults.
Might y'all already have an in house conversion script to go from the more recent meds_db format (via FEMR 0.2.4 / meds 0.3.3 / meds-reader 0.1.16 / meds-etl 0.3.12) from meds_reader to the PatientDatabase format compatible with MOTOR in FEMR 0.1.16?
Similarly, CLMBR-t-base requires meds 0.1.3 and CLMBR-gpt from hf_ehr requires yet another format.
Do y'all have suggestions or tools for navigating all these different patient database formats?
Much appreciated!
@xiubert Transforming between patient database formats wouldn't really help. These models / code / ETL versions also have different architectures/tokenization/ETL transformation variations that make them incompatible with each other. You might be able to get the code to run, but you likely wouldn't get good results since you would be providing data in a format that the model is not expecting.
I would only use them with the exact code and ETL versions specified in the documentation for each model.