v0.5.1

🐛 Bugfix:
Error in medication processing of eICU, see #32

v0.5.0

⚡Speedups:

• Support for time resampling, min-max clipping, and pivoting data to wide format is temporarily dropped. They will be moved later in the pipeline, this provided major speedups.
• Exploiting polars laziness to provide fast harmonization and low memory pressure.
• No more processing by patient chunk and individual patient files.

The full OMOP-ization pipeline can be run for all 5 databases in a single day :)

🐛bugfix:

• ICU stays with missing Length-of-stay data were dropped from the database. All patients are now preserved.
• Drug exposures that were not omop-ized were kept, with drug_concept_id = 0, previously they were dropped.

🏃 Getting further:

• Drug dosages were partially omop-ized: dosage and routes were extracted. Some units were omop-ized, routes were not harmonized yet.
• Observation period table was added
• drug strength table is still a work in progress, contributions are welcome ! especially for eICU.

v0.4.2

Changes :
⚡ Speedup : Converting MIMIC-III, MIMIC-IV and Amsterdam's csv.gz files as parquet in step 1. This conversion is only done once and allowed speeding up the following step.

v0.4.1

Changes :

• 🐛 Bugfix : visit_occurrence_id is no longer missing from condition_occurrence table.
• ⚡ Speedup : Converting eICU's csv.gz files as parquet in step 1. This makes re-running 1_extract_eicu.py 3 times faster.

v0.4.0

Started to speed up some operations using polars.

v0.3.2

Corrections on variables and dtypes in final OMOP tables.

Bugfix:

• Removed visit_start_date from measurement table, and string values in care_site table's place_of_service_concept_id
• Save all OMOP tables to parquet + corrected wrong dtypes on some tables.
• Rounding times to the second. This avoids an error due to high precision in time when writing some records to parquet OMOP tables.

Minor changes:

• Refactored timeseriespreprocessing to timeseriesprocessor
• Option to skip reset_dir() when starting 2_{dataset}.py

v0.3.1

Major changes:

• Generated a numeric patient id for OMOP-standardization. (Issue #15 )
• Added some insight for running times of each scripts. (as suggested in Issue #24 )
• Simplified the structure of paths.json
• Fixed inconsistency in datetimes of OMOP tables : some datetime columns contained the date, other contained the time of day. Now they all contain the full datetime. Issue #26

Minor changes:

• Added unit_concept_id to auxillary_files/user_input/timeseries_variables.csv
• Fixed harmless SettingWithCopy warning happening in database_processing/dataprocessor.py

Thanks to @mostafaalishahi, and @xinyuejohn for their contribution to the project.

v0.3.0

What's changed

• BlendedICU now handles the latest version of MIMIC-IV. (Issue #1 )
• Timeseries files are saved in a way that reduces indexing times.
• update to package versions
• other minor corrections. (Issue #21 )

v0.2.2

Fixed issues in the omop tables: A patient_id column had the wrong values, and some columns from the OMOP standard were missing. see Issue #17

v0.2.1

Quickfix. Bug introduced in the latest commit. See Issue #15