Database Reference
In-Depth Information
CDSSs provide medical practitioners with knowledge and patient-specific
information, intelligently filtered and presented at appropriate times, to
improve the delivery of care [20]. CDSSs differ from clinical guidelines
in that they are more data driven and require access to patient specific
data to help physicians in their decision making process. In theory, such
systems promise to offer customized and personalized decision support.
While several CDSSs are being reported in the literature, few of them
make full use of sensor data to assist in the decision making. In this
section, we focus on those clinical applications that do use sensor data
extensively to aid physicians in their decision making process. We sur-
vey applications in intensive care, operating rooms and in general clinical
settings.
3.1.1 Intensive Care Data Mining.
Today, critically ill pa-
tients are often attached to large numbers of body sensors connected
to sophisticated monitoring devices producing large volumes of physio-
logical data. Intensive care units are good examples of such data- rich
environments where multiple streams of continuous data are typically
produced, on a per patient basis. These data streams originate from
medical devices that include electrocardiogram, pulse oximetry, elec-
troencephalogram, and ventilators, resulting in several kilobits of data
each second. While these monitoring systems aim at improving patient
care and staff productivity, they clearly have introduced a data explosion
problem. In fact, the vast majority of data collected by these monitor-
ing systems in Intensive Care Units (ICUs) is transient. In talking with
medical professionals, we learned that the typical practice in ICUs is for
a nurse to eyeball representative readings and record summaries of these
readings in the patient record once every 30-60 minutes. The rest of the
data remains on the device for 72-96 hours (depending on the memory
capacity of the device) before it times out and is lost forever. Hospitals
are simply not equipped with the right tools to cope with most of the
data collected on their patients, prompting many to state that medical
institutions are data rich but information poor.
The potential of data mining in this area has been recognized by many.
Several efforts are underway to develop systems and analytics able for the
modeling of patient states and the early detection of complications. In
general, early detection of complications can lead to earlier interventions
or prophylactic strategies to improve patient outcomes. Early detection
rests on the ability to extract subtle yet clinically meaningful correlations
that are often buried within several multi-modal data streams and static
patient information, spanning long periods of time.
