Anesthesia depth monitor
During my intership at NTT Data Institute of Management Consulting, I conducted field observations and developed low-fidelity UI proposals for a new device to help surgeons monitor anesthesia depth.
A major problem encountered in endoscopies is maintaining the depth of anesthesia. Too shallow, and the patients feel the pain or leave the operating room with a traumatic memory, but too deep and the patient stops breathing, a situation that is sometimes fatal. There are three aspects of anesthesia we needed to address in this particular case - pain, consciousness, and physical sedation. In this project, we wanted to specifically measure the consciousness aspect, which is associated with patients having memories of the procedure. Surgeons have traditionally measured it by observing patient movements or expressions of pain. However, not only is this subjective, but even if the patient was in pain or became conscious during the procedure, they may not be able to express it due to their physical sedation. In the worst case, the patient would be aware and in pain but would be unable to communicate it, resulting in a traumatic experience.
We designed an in-house EEG device shaped like an earphone targeted primarily for consumers. In this project, our goal was to evaluate the device's potential for application in monitoring patient consciousness levels during endoscopies. We focused on endoscopies because they are regularly performed, are often scheduled in advance, and are relatively low risk, while still facing anesthesia depth as a major challenge. We created an individualized model of patients' personal consciousness models by using the first 3 minutes of the procedure to gather their EEG data while they were still conscious. We used ML on this data to have the program learn the patterns of the patient's brain waves while they were consciousness, so that if they were approaching consciousness during the procedure, the model can identify that pattern.
Endoscopies are often carried out by a small team, with a few nurses and one doctor who performs both the endoscopy and manages the anesthesia levels at the same time. Having a device to monitor anesthesia depth can help doctors by acting as an additional observer of the patient.
Doctors/surgeons: The movements from a patient struggling or fighting while they are under the anesthesia are dangerous, and can cause the surgeon to make unintentional cuts. By knowing that the patient is starting to wake up in advance, they can give additional dosages of anesthetic medication before the patients start to regain consciousness and move.
Patients: Anesthesia depth needs to be maintained; not too deep, not too shallow. Leaving the operating room with traumatic memories can lead to aversions of future procedures and PTSD, and should be avoided. However, when it's too deep, this can cause respiratory arrest that can lead to death. In addition, there are evidence to suggest that anesthesia depth is associated with recovery time, and postoperative cognitive functions . These are some of the reasons why an accurate monitoring system is important to patients.
Field studies were conducted during 12 scheduled endoscopies between October and December 2022, at the National Cancer Center Japan, Kashiwa Campus. 2 of the endoscopies were used for pilot trials, while we conducted data collection in the last 10.
One existing monitoring device that uses EEG is called the BIS monitor, which has many challenges including that it is a black box model. We used Richmond Agitation-Sedation Scale (RASS) as the baseline value to measure consciousness without using EEG. To see whether our device was able to make better estimations than the existing BIS device, we compared the device-RASS correlation to BIS-RASS correlation.
As a UX designer/researcher, my role was to attend the endoscopies in person, make observations on how the surgeons worked in the surgery room, and create proposals of how we can relay information from the device to the surgeons. I conducted short hearings from the 4 surgeons involved in the project to incorporate their opinions. Additionally, I was responsible for the data collection, recording anesthesia administration and dosages, RASS values, and all other major events during the surgery.
During the procedures, we ran the consciousness estimation program to get a feel for how the program behaved in the actual usage environment. Given the large interpersonal differences in expression of EEG waves, which is a notorious challenge for these devices, my supervisor and I ran the consciousness estimation program based on the personal model of the patient, and the general population model which combined data of multiple patients from previous studies
Effectiveness of our models
In terms of functionality, there was a significantly stronger correlation between the predicted consciousness and the RASS values for our system than for the existing BIS device (p = 0.001). This suggests that our device has promising potentials and further tests should be conducted.
Contrary to our prediction, the general model performed better than the personalized model. Thus, we made the decision to only include the general model in the main interface of the new proposed device, and having the personal model as an additional option in cases where the general model seemed to not be working. One potential explanation for this may be that the 3 minutes of EEG data at the beginning of the operation was not enough to create an accurate model of the individual patient.
As for how the surgeons operated in their environment, one big finding was the most of their visual attention was focused on the endoscopy monitor. Thus it is important for us to incorporate other senses such as sound when communicating the information back to the surgeons.
Surgeons wanted to see the data and predictions from the models, so that they had more control over how they applied the information to in their actions and decisions.
Another important point to address was that surgeons will not have the time to interact with our device and figure out information while they are operating. Thus, any information needs to be communicated in a clear and concise manner, so that they can be understood very quickly.
After getting feedback from the surgeons, it was agreed that it would be helpful to have a physical display or a monitor they could take quick looks at. Considering the cost and feasibility, my supervisor and I decided creating an app and running it on an off-the-shelf device such as an iPad would be as realistic solution. Below are some of the suggested placements for where this monitor could sit with respect to the main endoscopy monitor they have, as well as where LED strips can be placed if we chose communication by color.
The below are some of the ideas I came up with on what we should include in the main display:
For the main monitoring screen, we decided to show the anesthesia depth predictions from the general model only, as the personal model wasn't as accurate as we expected. We included a chronological graph of the predicted consciousness level from the start of the procedure to provide a point of reference. In addition, we wanted to show the connection status of the EEG device on the patient's scalp, as well as "one-ear mode" which can be deployed if the connection from one electrode was unstable.
The next sketches are on how we can display these information on the main monitor:
In terms of methods of communicating the information about anesthesia depth, I considered using sound, light flashes, and colors. Below are some proposed ideas.
Challenges to address in future studies
My supervisor, the doctor form the National Cancer Center that was involved in this project, and I identified some challenges that can be addressed in future iterations:
Multiple kinds of interpersonal differences: in EEG expression due to multiple demographic factors including age, pain tolerance, in sensitivity to anesthetics, in the pain experienced while the electrocauter is on vs when it's off.
We used one anesthetic in this study, but we will need data on the other drugs for this to become applicable for more cases.
The user needs will be different for every facility and every doctor, which are influenced by multiple factors including the number of personnel on the operating team, and personal preference.
We will need to assess if and how the doctors behaviours change when they use the device. We want to ensure that the information is helpful, and not detrimental in the decision making process. For example, we don't want a case where the doctor administered additional anesthetic dosages based on what our device said, only for it to be too much and fatal for the patient.
 Farag, E., Chelune, G. J., Schubert, A., & Mascha, E. J. (2006). Is depth of anesthesia, as assessed by the bispectral index, related to postoperative cognitive dysfunction and recovery? Anesthesia & Analgesia, 103(3), 633–640. https://doi.org/10.1213/01.ane.0000228870.48028.b5