The MIMIC II Clinical Database (Public Subset)

This database is described in

GB Moody, LH Lehman. Predicting Acute Hypotensive Episodes: The 10th Annual PhysioNet/Computers in Cardiology Challenge. Computers in Cardiology 36:541-544 (2009).

Please cite this publication when referencing this material, and also include the standard citation for PhysioNet:

Goldberger AL, Amaral LAN, Glass L, Hausdorff JM, Ivanov PCh, Mark RG, Mietus JE, Moody GB, Peng C-K, Stanley HE. PhysioBank, PhysioToolkit, and PhysioNet: Components of a New Research Resource for Complex Physiologic Signals. Circulation 101(23):e215-e220 [Circulation Electronic Pages;]; 2000 (June 13).

The non-text portions of 110 records from the MIMIC II Clinical Database were made available as parts of the training and test data sets for the PhysioNet/Computers in Cardiology Challenge 2009.

Each record in the MIMIC II Clinical Database contains information about a single patient. About 11% of the patients have been admitted to one or more of our study ICUs more than once, and their records contain information about all of these admissions, supplemented by limited additional information collected during the intervals between ICU admissions. (In the MIMIC II Waveform Database, however, each record contains information for only a single ICU admission.)

Each record in the public subset is contained within a subdirectory of this one. By visiting any of the numbered directories listed below, it is possible to get a view of the record in chronological order, and this may be useful to obtain an overview of the record. The same information is contained in a WFDB annotation file within each directory, and also in a plain text file, and either of these formats will be more easily parsed by a computer program than the default (HTML) view.

Many of the timestamps that appear in the MIMIC II Clinical Database were manually entered into the ICU medical information systems as much as an hour after the events to which they refer, and others were extracted from laboratory reports that became available at varying times following the test. In such cases, the MIMIC II Clinical Database includes both the time of data entry (the time at which the information was first available in the medical record) and the best available estimate of the time of the event or test itself. These times are not perfectly synchronized with those in the corresponding MIMIC II Waveform Database records, as a result of imperfect synchronization of clocks as well as imprecise or faulty recall, and occasional transcription errors in manual entry. As a result, one should not assume that the time order of events separated by 15 minutes or less is known; for example, an intervention recorded in the clinical database may appear a few minutes earlier than the event in the waveform database that prompted the intervention.

First-time users of the MIMIC II Database will find much useful information in our tutorial, Using the MIMIC II Database.