In light of there is no offcial guide of deploying MIMIC-IV to Postgres server, this repo will help you navigate this process. We made necessary modification based on the offcial guide of MIMIC-III to adapt to MIMIC-IV.
Complete the PhysioNet Credentialing and download MIMIC-IV data. Decompress all .csv.gz file using "gzip -d *.gz". You will have such directory structure:
BASE_DIR
│
└───core
│ │ admissions.csv
│ │ patients.csv
│ │ transfers.csv
└───hosp
│ │ d_hcpcs.csv
│ │ ...
└───icu
│ │ d_items.csv
│ │ ...
Please refer here for postgresql download and installation
In my case, I create a user named mimicuser, a new database named mimic4 containing a new schema named mimiciv holding all data by:
createuser -P -s -e -d mimicuser #creat user
psql -U mimicuser #launch psql
CREATE DATABASE mimic4 OWNER mimicuser; #create new database
\c mimic4; #enter the new database
CREATE SCHEMA mimiciv; #create new schema
\q #quit psqlpsql 'dbname=mimic4 user=mimicuser options=--search_path=mimiciv' -f create_tables.sqlChange /YOUR/BASE/DIR/ to your BASE_DIR in the line 3,11,27 and run:
psql 'dbname=mimic4 user=mimicuser options=--search_path=mimiciv' -f load_data.sql- The data description and real data are inconsistent in data format, varchar length, etc. This repo is implemented based on the real data. Some comments are left in create_tables.sql to mark down the difference.
- We follow the installation of MIMIC-III that creating 10 subtables for labevents.csv and chartevents.csv, respectively. Chartevents table is divided by itemid to 10-fold, while labevents table is divided by labevents_id.
- The size of MIMIC-IV after de-compression may exceed 100GB. Therefore, loading data is time-consuming. You may expect several hours to run through the process.
- This repo is based on the latest data release before Aug 21, 2020. Should there be any data update, incompatibiltiy may observed.
- Leave an issue, if you encounter any error.