This video examines details of the MIMIC-III data model, which is the clinical data model that we use throughout this course and specialization. MIMIC stands for Medical Information Marked for Intensive Care, and is part of a larger data project called PhysioNet, which is a large open source collection of physiologic and clinical data, submitted by many institutions. MIMIC, developed by the MIT Lab for computational physiology, is a data set that is derived from patients admitted to the Intensive Care Unit at Beth Israel Deaconness Medical Center, in Boston Massachusetts. The full data set is quite large, consisting of over 53 thousand hospital admissions over an 11-year time frame. The data set, includes patients admitted to medical, surgical, pediatric, and neonatal ICUs at Beth Israel. The full data set is available at no cost, but requires specific certifications and licensing agreements, in order to access it. The MIMIC team has released a 100-patients subset into the public domain, that does not have these additional requirements. This is the data set that we use for this class. This diagram shows the breadth of the data sources and data types that the MIMIC team has collected, including structured data elements, such as patient demographics, diagnoses, procedures, and medications, plus unstructured clinical notes and physiological signals. The data set is processed to remove all patient identifiers. Dates are replaced with fictitious future dates, and data are organized into the MIMIC-III data model structure. Here, you can see the expansive coverage of the patient population in the full MIMIC-III data set. Later, we will assess how well the smaller 100-patient data set reflects the patient population characteristics, of the full MIMIC-III data set. Details on every table in MIMIC are available from the mimic website, mimic.physionetnet.org. This URL is also available in the course readings. On the left navigation bar that appears on the mimic homepage, select tables in MIMIC topic. This sub-menu, lists every table available in MIMIC-III in alphabetical order. There are 26 tables in the MIMIC-III data model. Selecting a table brings up a page that provides general information about the contents of the table, such as the source of data in the table, a short description of the data, the number of rows in the full data set, and how the table links to other tables. In this example, I have selected the admission table. Note that the admissions table links to the patient's table, using the common subject ID, key field. This information is used to construct SQL-joined statements, that involve two or more tables. Following a brief summary and a specific note, details are provided about the data stored in the fields in the table. On the right side are buttons that navigate to specific areas of the data dictionary documentation for the admissions table, topics related to overall summary features appear in red text. Data definitions for specific columns, appear in blue text. Of note, MIMIC does not provide explanations for every column in the table. For example, the admissions table has detailed data definitions for 15 columns, but the actual admissions table contains only 19 columns. You can usually determine the meaning of the remaining columns, by the column name or by the data values, contained in the column. The MIMIC documentation, also provides information about the clinical data system that was used to source the data values in MIMIC. This information, called metadata, is useful for understanding how the data are collected, and why data from these sources, may look different than similar data collected from other clinical systems. The MIMIC data model is relatively simple to understand and query, since there is consistent naming of common linkage columns for joint conditions across tables. For example, values in the column subject ID, always refer to a specific patient, no matter in what table that column appears. Thus, subject ID can be used in joints to connect information about individual subjects across two or more tables in MIMIC.
