Database Reference
In-Depth Information
The success of variability analysis has been reported by many re-
searchers. In [51], the authors perform a spectral analysis of heart rate
measurements to show a relationship between heart rate variability and
sepsis. In [47], the potential of this approach is highlighted with a de-
scription of multiple clinical applications that use such complexity anal-
ysis. In [52] the authors derive several empirical links between heart rate
variability and mortality in intensive care units. In [53], the prognostic
potential of heart rate variability measures in intensive care is proposed.
The authors in [54] have shown that reductions of heart rate variability
are correlated with outcomes in pediatric intensive care. At the Univer-
sity of Virginia, Lake [45] et. al. have used the sample entropy on heart
rate measurements to predict the onset of sepsis in neonates. In [55] the
predictive capability of heart rate variability on the prognosis of a large
population of trauma patients is described.
Heart rate variability has also been use to determine when to extu-
bate or remove patients from mechanical ventilation in intensive care
[56]. A clinical trial is currently underway in Canada testing whether
maintaining stable heart rate and respiratory rate variability through-
out the spontaneous breathing trials, administered to patients before
extubation, may predict subsequent successful extubation [56].
Besides heart rate variability analysis, there are many other applica-
tions of sensor data mining intensive care. Analysis of the dynamics of
the ECG signal has enabled researchers to build systems for arrhythmia
detection using standard machine learning and classification techniques.
The work presented in [57] is illustrative of these systems.
Respiratory complications have also received a significant amount
of attention in the intensive care community. In [58] the authors de-
scribe the use of sensor data from brain activity measured with elec-
troencephalograms (EEG), eye movements measured with electroocu-
logram (EOG), muscle activity measured with electromyogram (EMG)
and heart rhythm measured with ECGs during sleep, to detect obstruc-
tive sleep apnea episodes, that are known to be correlated with poor
patient outcomes.
EEG signals have also been used beyond sleep apnea studies. In [59],
EEG spectral analysis is performed to detect epileptic seizures with ma-
chine learning techniques, while in [60], continuous EEG spectral anal-
ysis for brain ischemia prediction is illustrated.
General predictive models for patient instability in intensive care have
also been proposed in the literature. A notable example is the work in
[61], where the authors extract several time series trending features from
heart rate and blood pressure measurements collected every minute and
build predictive models using a multi-variable logistic regression mod-
