Software Open Access
Cardiac Output Estimation from Arterial Blood Pressure Waveforms
Published: Aug. 2, 2007. Version: 1.0.0
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 (2003). Circulation. 101(23):e215-e220.
Cardiac output (CO), defined as the volume of blood pumped by the heart per unit time (often expressed in liters per minute), is the critical variable characterizing circulatory function, but it is also one of the most difficult to measure. Thermodilution cardiac output (TCO) is generally regarded as the "gold standard" among CO measurements. By introducing a small amount of cool fluid into the blood upstream of the heart, and observing how rapidly the blood temperature equilibrates, the flow and thus the CO can be calculated. TCO is highly invasive, intermittent, difficult to perform, and has sufficient potential to cause complications that it is used only when an accurate CO measurement is essential -- and even then, TCO errors are on the order of 15 to 20%.
For over a century, many researchers have proposed alternative, minimally invasive or non-invasive methods of estimating CO, based on features of the arterial blood pressure waveform.
This project contains the code used for a study that rigorously compares 11 estimates against TCO using 120 subjects from the MIMIC II database (all subjects for whom multiple TCO measurements were recorded at the time of the study).
The code directory contains Matlab implementations of the CO estimation algorithms, together with additional Matlab code to provide a set of tools for exploring CO estimators.
The software in code/1extract/ can use the waveform and numerics records to generate input files for the remaining software.
The data directory contains a list of the original MIMIC II records used in this study.
At the time of the 2005 study by Sun et al., the MIMIC II Waveform Database included 120 pairs of records (the "study records") that included both ABP waveforms and TCO measurements, and all of these were used in the study. The ABP signals, sampled at 125 Hz, can be found in the waveform records, and the reference TCO measurements, occurring at irregular intervals, can be found in the companion "numerics" records (identified by the same record names, but with an appended "n"). Demographic and clinical information, including age, gender, ICD9 codes, and medications, was available in the MIMIC II Clinical Database for 117 of the study records.
- The file RECORDS lists the study records, identifying them by their names within the now-obsolete version 2 of the MIMIC II Waveform Database, as was used in the study. If you wish to do anything other than exact replication of the 2005 study, we recommend using the corrected and expanded MIMIC II Waveform Database Matched Subset (for which age, gender, and detailed clinical information are available) or the much larger MIMIC II Waveform Database, version 3 (which includes many additional records for which age, gender, and clinical information are not available).
- File RECORDS-2005 also contains the names of the corrected copies of the study records within the current MIMIC II Waveform Database, version 3, as well as their version 2 record names, and the subject IDs for the associated patients' records in the MIMIC II Clinical Database.
- File RECORDS-2014 contains an updated list of 535 records from 350 subjects with ABP signals, TCO measurements, age, and gender in the MIMIC II Waveform Database Matched Subset, and clinical data in the MIMIC II Clinical Database, including the 117 study records for which demographic and clinical data are available.
The first column of RECORDS-2005 contains MIMIC II subject IDs for each record, which can be used to associate waveform and numerics data with demographic and clinical data. For example, in the MIMIC II Waveform Database Matched Subset, the ABP signals and TCO measurements for subject ID "s00020" can be found within http://physionet.org/physiobank/database/mimic2wdb/matched/s00020/. The ABP and other signals are in waveform record s00020-2567-03-30-17-47 (as can be determined at the link above), and the TCO measurements are the non-zero values of the CO variable in numerics record s00020-2567-03-30-17-47n. The fields appended to the subject ID in these record name are the surrogate date (2567 March 30) and the actual starting time of the record (17:47 in the local time zone where the recording was made).
In the MIMIC II Clinical Database, the initial "s" and any leading zeros are omitted from the subject ID. For example, demographic and clinical data for "s00020" appears in the MIMIC II Clinical Database as subject ID "20". The last three cases in RECORDS-2005 are shown as subject ID "sxxxxx"; in these cases, there are no corresponding records in the MIMIC II Clinical Database, and age, gender, ICD9 codes, and medications are unknown for these three subjects.
The second column of RECORDS-2005 contains the record names from the (obsolete) MIMIC II Waveform Database, version 2.
For studies that require using the uncorrected data used in the 2005 study, within http://physionet.org/physiobank/database/mimic2db/ are subdirectories for each record, containing WFDB-compatible waveform and numerics records. For example, the first entry in RECORDS is a40006; find the ABP signal in record mimic2db/a40006/a40006 [layout header], and the TCO measurements in record mimic2db/a40006/a40006n [header].
The third column of RECORDS-2005, and any additional columns, contain the record names from the MIMIC II Waveform Database, version 3. The contents of these records are identical to those in the MIMIC II Waveform Database Matched Subset, except that their .hea files are undated (they contain the starting times but neither the actual nor the surrogate dates, and they do not contain the subjects' age and gender). Note that if there is a gap of an hour or more in a subject's waveform and numerics recordings, there will be more than one MIMIC II Waveform Database version 3 record for that subject; although TCO measurements will be available in at least one of the records for each subject, they may not be present in all of a given subject's records. See the main page for the MIMIC II Waveform Database, version 3, for details on locating these records.
The other files here contain additional clinical information extracted from the MIMIC II Clinical Database for 117 of the study records:
- demographics.txt: sex, age*, and mean and standard deviation of hematocrit. * age over 89 is given as >=90
- icd9.txt: ICD9 codes and brief descriptions during the ICU stay
- medications.txt: intravenous medications administered during the ICU stay
(not necessarily while ABP was being monitored)
Documentation and References
Detailed descriptions of the algorithms and data collected here can be found in the following papers, contained in the doc directory:
- Sun JX. Cardiac Output Estimation using Arterial Blood Pressure Waveforms. [MEng thesis]. Cambridge, MA: Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, September 2006.
- Sun JX, Reisner AT, Saeed M, Mark RG. Estimating Cardiac Output from Arterial Blood Pressure Waveforms: a Critical Evaluation using the MIMIC II Database. Computers in Cardiology 32:295-298 (2005).
Anyone can access the files, as long as they conform to the terms of the specified license.
License (for files):
Open Data Commons Attribution License v1.0
Total uncompressed size: 0 B.