Senior Design: showing mastery of undergraduate skills in Capstone
Developed during the University of Pittsburgh courses, BIOENG 1160: and 1161 Bioengineering Design, this team project focuses on unmet need identification and collaboration. Through this iterative design process, a device was developed to mitigate alarm fatigue.
Relevant skills: ethnography, design documentation, MATLAB, Simulink, IFTTT, FDA considerations
Learn more
This process began with human-centered empathy-based consideration of hindrances in the medical field and the feasibility of finding a solution in a year-long, COVID-era project. A member of the team, a nursing student through the University of Pittsburgh, described the toll of alarm fatigue--the process of a high number of alerts (often false alarms) on hospital floors desensitizing nurses to the sound and increasing their response time or causing them to miss actual emergencies. Research into the topic revealed current solutions focus on vigilance training or services that offer consultation and reconfiguring of the entire alarm system. Further discussion with practicing nurses and hospital professionals revealed the need for a solution to promote accurate awareness of alarm occurrence and reduce widespread alarm fatigue while staying unobtrusive to the administration of medical care, accounting for the wide range traveled by some nurses on a floor, cleanable, durable, and not taking a large amount of time to implement, install, or test the system is working properly.
Pictured left is the University of Pittsburgh School of Nursing Skills Lab, where my teammates and I were able to observe typical machinery in a hospital room, their alarms, and hear first hand accounts of alarm fatigue.
Through the development of minimum viable prototypes and killer experimentation, our team found that sound and vibration were the most effective alerts. Interviews with nursing students introduced us to the use of shift phones by area hospitals. This is a cellular device that sends a phone call to the nurse attending when if the nurse call button is pressed. While previous ideas had centered around wearable cuffs, necklaces, and earpieces that could transmit personalized alarms--user feedback guided us towards making out device also capable of ringing a shift phone.
Our team first chose an Alaris Infusion Pump as the alarm we would focus on mitigating fatigue for. The pump is a high contributor to alarm fatigue as it makes the same sound for several notifications such as end of treatment, air-in-line, or obstruction of flow. Using a Raspberry Pi microprocessor running Simulink code, we developed our first iteration prototype: a device capable of reading live sound, detecting the Alaris alarm through a matched filter, and using the Rasberry Pi API to trigger a phone call using a third party platform. This served as a supplement to the existing alarm pathway but it's high false alarm rate meant we still had much more work to do.
While I served as the primary coordinator for our team, my main technical role was assisting in the development of the detecting program. Our code went through eight iterations, from a matched filter to two programs that worked in tandem. One MATLAB function that identified the three most prominent frequencies of the alarm from a stagnant clip and a second which detected those three tones from live sound. If the output of the detect function superseded a threshold indicative of a 95% confidence level, the program output the necessary signal to a Raspberry Pi webhook to initiate the phone call through If This Than That an internet-based applet capable of sending phonecalls.
In terms of the hardware: the program was run on a Model 3B+ Raspberry Pi microprocessor with USB mini microphone. The housing was an acryllic case with a custom 3D-printed clip made of HTPLA that allowed it to be slotted into a standard IV pole clamp. Due to this device being independent of the Alaris pump, it did not constitute a medical device. However, it or a feature similar to it could be incorporated into future iterations of current or novel medical devices. As part of the design process, I explored relevant FDA regulatory pathways and wrote a hypothetical Pre-Sub to evaluate test protocols and technical characteristics as if the device were following a Special 510k pathway as a new feature to an existing device.
Preliminary verification and validation testing occurred. Including work with nursing students at the Winter Institute for Simulation, Education, and Research. We found our device to be promising in terms of range (exceeding 200ft), durability, and within time constraints for identifying alarms. However, when performing in an environment simulating a hospital floor, the device failed to improve response time and was found to be difficult to clean. The lack of cleanability was determined to be due to the structure of the 3D-printed clip and was revised. The phone call had failed to trigger or triggered slowly on multiple tests. Based on review of device output, it was determined that the Raspberry Pi had detected the sound but had failed to trigger the webhook or the third-party applet had failed to respond or responded slowly. Based on feedback from users, the relative level of noise, complexity and nature of task, and proximity to the alarm were unrealistic to a hospital scenario and failed to induce alarm fatigue--meaning, in this testing, our device could not outperform the natural abilities of the nurses.
Moving forward, both our device design and testing protocol need to be improved. This project will not continue. If it were, future steps would include to finding a more reliable third-party to trigger the phone call or acquiring the ability to perform this function independently. Further, future validation testing should increase the intensity of background noise, have a more complex task that needs to be completed on a deadline, and find an opportunity to hear from a larger number of nursing students but also professional nurses in the field.
My role on this team tends to be that of a coordinator, documenter, and facilitator: as such, I have had a hand in developing a large breadth of this project, though perhaps not in the same depth as some of my teammates who have focused on specific aspects. As it stands, this project failed to adequately address alarm fatigue but reinforced the importance of letting end user feedback drive the design process.
