Database Reference
In-Depth Information
Systems for data mining in intensive care:
Modern patient monitors have evolved into complex system that not
only measure physiological signals but also produce alerts when the phys-
iological state of the patient appears to be out of range. State of the art
patient monitors allow physicians to program thresholds defining nor-
mality ranges for physiological systems. For example, one can program
a patient monitor to produce an audible alert if the oxygen saturation
level of the blood is below 85 percent. The values of these thresholds are
typically obtained from general guidelines or from data mining processes.
Such simple alerting schemes are well known to produce very large num-
bers of false alarms. In [23], it is reported that more than 92 percent of
alarms generated in an ICU are of no consequence. Furthermore, there
are many complex physiological patterns of interest to physicians that
cannot be represented by a set of thresholds on sensor data streams.
Several research initiatives are addressing this problem with the design
of platforms facilitating analysis beyond the simple thresholding capa-
bilities of existing patient monitoring systems.
One example is BioStream [24], a system that performs real-time
processing and analysis of physiological streams on a general purpose
streaming infrastructure. The authors use ECG data along with tem-
perature, oxygen saturation, blood pressure and glucose levels as inputs
into patient-specific analytic applications. The system supports a dif-
ferent processing graph (for analysis) per patient, where the graph can
be composed of system supplied operators (functions) and user imple-
mented operators. The authors also state that BioStreams can be used
to discover new patterns and hypotheses from the data and test them,
however there is limited discussion of the underlying analytics and use
cases.
In [25] the authors describe an architecture for a system whose goals
are data mining, fusion, and management of data streams for intensive
care patients. The proposed system has online components for capture
of physiological data streams and program execution along with off-line
components for data mining.
The SIMON (Signal Interpretation and MONitoring) platform [26]
developed at Vanderbilt is a data acquisition system that continuously
collects and processes bedside patient monitoring data. SIMON collects
typical ICU monitoring vital signs including heart rate, blood pressures,
oxygen saturations, intracranial and cerebral perfusion pressures, and
EKG-ECG waveforms. This data collection is intended to support clini-
cal research by enabling further analysis and mining. The system is also
