MIMIC Primer 201: The Scope of MIMIC

By: Ned McCague
Data Scientist at Kyruus

The MIMIC Database is quite literally one of the best databases in the world. Within it, you can find hundreds of tables and thousands of fields that pertain to specific inpatient stays within an Intensive Care Unit (ICU). With over 40,000 ICU stays, this provides an unbelievable amount of data with impressive potential and applicability in multiple arenas. You have access to medication data, biometric data, billing data (i.e. ICD-9 codes, DRG codes, and CPT codes), and event data. You can even take certain data points and create time-to-event data to explore the data in ways that haven’t been looked at before.

But, with so much data, it can be easy to get lost in the enormity of it all. People who get access to the data often find themselves saying “There are so many tables! I don’t even know where to start!”

So let’s simplify things. With just a few tables, you can actually cover a lot of ground and gain access and insights into the various features of the data within the MIMIC Database. Which tables? Personally, I would focus on three tables: (1) ICU Stay Detail, (2) Chart Events, and (3) Lab Events. They’re probably the most important and commonly used tables in the MIMIC Database, and they’re likely to give you the most bang for your buck.

The ICU Stay Detail table has data pertaining to a unique admission to the ICU. This includes things like a patient’s gender, date of birth, date of admission to the hospital, date of admission to the ICU, date of death, and length of stay.
The Chart Events table contains data on the vital signs that get recorded by clinical staff. This includes blood pressure, body temperature, heart rate, and many more.
Finally, the Lab Events table contains data about the specific lab tests that were performed. So if you want information on something like a patient’s serum glucose levels, white blood cell counts, or serum potassium levels, you’d want to use this table.

And of course, all of the data is scrubbed, aggregated, and deidentified to be totally HIPAA compliant.

For a walk through of how to pull some of this data using SQL, watch the video here: http://criticaldata.mit.edu/mimic-ii/