Georgia Tech researchers Wassim Haddad, Allen Tannenbaum and Behnood Gholami (left-right) and Northeast Georgia Medical Center chief medical informatics officer James Bailey have developed control algorithms to automate sedation in hospital intensive care units. Their algorithms use clinical data to accurately determine a patient’s level of sedation and can notify medical staff if there is a change in the level. Credit: Gary Meek |
Researchers at the Georgia Institute of Technology and the Northeast Georgia Medical
Center are one step
closer to their goal of automating the management of sedation in hospital
intensive care units (ICUs). They have developed control algorithms that use
clinical data to accurately determine a patient’s level of sedation and can
notify medical staff if there is a change in the level.
“ICU nurses have one of the most task-laden jobs in medicine
and typically take care of multiple patients at the same time, so if we can use
control system technology to automate the task of sedation, patient safety will
be enhanced and drug delivery will improve in the ICU,” said James Bailey, the
chief medical informatics officer at the Northeast Georgia Medical Center in
Gainesville, Ga. Bailey is also a certified anesthesiologist and intensive care
specialist.
During a presentation at the IEEE Conference on Decision and
Control, the researchers reported on their analysis of more than 15,000
clinical measurements from 366 ICU patients they classified as “agitated” or
“not agitated.” Agitation is a measure of the level of patient sedation. The
algorithm returned the same results as the assessment by hospital staff 92% of
the time.
“Manual sedation control can be tedious, imprecise,
time-consuming and sometimes of poor quality, depending on the skills and
judgment of the ICU nurse,” said Wassim Haddad, a professor in the Georgia Tech
School of Aerospace Engineering. “Ultimately, we envision an automated system
in which the ICU nurse evaluates the ICU patient, enters the patient’s sedation
level into a controller, which then adjusts the sedative dosing regimen to
maintain sedation at the desired level by continuously collecting and analyzing
quantitative clinical data on the patient.”
This project is supported in part by the U.S. Army. On the
battlefield, military physicians sometimes face demanding critical care
situations and the use of advanced control technologies is essential for
extending the capabilities of the health care system to handle large numbers of
injured soldiers.
Working with Haddad and Bailey on this project are Allen
Tannenbaum and Behnood Gholami. Tannenbaum holds a joint appointment as the
Julian Hightower Chair in the Georgia Tech School of Electrical and Computer
Engineering and the Wallace H. Coulter Department of Biomedical Engineering at
Georgia Tech and Emory
Univ., while Gholami is
currently a postdoctoral fellow in the Georgia Tech School of Electrical and
Computer Engineering.
This research builds on Haddad and Bailey’s previous work
automating anesthesia in hospital operating rooms. The adaptive control
algorithms developed by Haddad and Bailey control the infusion of an anesthetic
drug agent in order to maintain a desired constant level of depth of anesthesia
during surgery in the operating room. Clinical trial results that will be
published in IEEE Transactions on Control Systems Technology
demonstrate excellent regulation of unconsciousness allowing for a safe and
effective administration of an anesthetic agent.
Critically ill patients in the ICU frequently require
invasive monitoring and other support that can lead to anxiety, agitation and
pain. Sedation is essential for the comfort and safety of these patients.
“The challenge in developing closed-loop control systems for
sedating critically ill patients is finding the appropriate performance
variable or variables that measure the level of sedation of a patient, in turn
allowing an automated controller to provide adequate sedation without
oversedation,” said Gholami.
Georgia Tech researchers Allen Tannenbaum, Wassim Haddad and Behnood Gholami (left-right) and Northeast Georgia Medical Center chief medical informatics officer James Bailey have developed control algorithms to automate sedation in the intensive care unit. Their algorithms returned the same results as the assessment by hospital staff 92% of the time. Credit: Gary Meek |
In the ICU, the researchers used information detailing each
patient’s facial expression, gross motor movement, response to a potentially
noxious stimulus, heart rate and blood pressure stability, noncardiac
sympathetic stability, and nonverbal pain scale to determine a level of
sedation.
The researchers classified the clinical data for each
variable into categories. For example, a patient’s facial expression was
categorized as “relaxed,” “grimacing and moaning,” or “grimacing and crying.” A
patient’s noncardiac sympathetic stability was classified as “warm and dry
skin,” “flushed and sweaty,” or “pale and sweaty.”
They also recorded each patient’s score on the motor
activity and assessment scale (MAAS), which is
used by clinicians to evaluate level of sedation on a scale of zero to six. In
the MAAS system, a score of zero represents an
“unresponsive patient,” three represents a “calm and cooperative patient,” and
six represents a “dangerously agitated patient.” The MAAS
score is subjective and can result in inconsistencies and variability in
sedation administration.
Using a Bayesian network, the researchers used the clinical
data to compute the probability that a patient was agitated. Twelve-thousand
measurements collected from patients admitted to the ICU at the Northeast Georgia Medical
Center between during a
one-year period were used to train the Bayesian network and the remaining 3,000
were used to test it.
In 18% of the test cases, the computer classified a patient
as “agitated” but the MAAS score described the
same patient as “not agitated.” In five percent of the test cases, the computer
classified a patient as “not agitated,” whereas the MAAS
score indicated “agitated.” These probabilities signify an 18% false-positive
rate and a five percent false-negative rate.
“This level of performance would allow a significant
reduction in the workload of the intensive care unit nurse, but it would in no
way replace the nurse as the ultimate judge of the adequacy of sedation,” said
Bailey. “However, by relieving the nurse of some of the work associated with
titration of sedation, it would allow the nurse to better focus on other
aspects of his or her demanding job.”
The researchers’ next step toward closed-loop control of
sedation in the ICU will be to continuously collect clinical data from ICU
patients in real time. Future work will involve the development of objective
techniques for assessing ICU sedation using movement, facial expression and
responsiveness to stimuli.
Digital imaging will be used to assess a patient’s facial
expression and also gross motor movement. In a study published in IEEE
Transactions on Biomedical Engineering, the researchers showed that
machine learning methods could be used to assess the level of pain in patients
using facial expressions.
“We will explore the relationship between the data we can
extract from these multiple sensors and the subjective clinical MAAS score,” said Haddad. “We will then use the knowledge
we have gained in developing feedback control algorithms for anesthesia dosage
levels in the operating room to develop an expert system to automate drug
dosage in the ICU.”