Analysis of Respiratory Kinematics Device: Breath Detection from Motion Sensor Signals; Racial Bias in the Case of the Optum Algorithm: How the Use of Healthcare Costs as a Label Introduces the Adversarial Actor Systemic Racism

My technical work and STS research both concern the adoption of computer science technologies in the health field. Through the combination of these projects, I examine the potential of computer science technologies to have positive and negative impacts on patients. My technical work involved developing algorithms for a respiratory monitoring device that is designed to enable earlier detection and intervention of labored breathing which would potentially improve patient outcomes. On the other hand, my STS research analyzes a widely used health care management algorithm called the Optum algorithm and how it leads to racial disparity in enrollment of patients to a high-risk care management program.
Entering a state of labored breathing is an important indicator of deterioration of patient respiratory function and a critical need for clinician intervention. Currently, labored breathing is identified by a bedside evaluation of visual cues such as blue colored mouth or fingernails and increased recruitment of accessory respiratory muscles to support the failing primary respiratory muscles (Romer & Polkey, 2008; "Signs of Respiratory Distress", n.d.). My capstone advisor had the idea to replace this bedside evaluation with a device consisting of motion sensors across a patient’s chest and back coupled with algorithms that could detect if a patient was entering labored breathing. A key step in creating the device is to separate breaths in the motion sensor data. My capstone team created an algorithm that identified landmark points in motion sensor signals that can be used to segment signals by breath. The goal of the technical work was to use computer science technologies to enable an analysis of what constitutes a normal vs labored breath.
The Optum algorithm, like the algorithms I designed for the ARK device, is a computer science technology created to better the patient experience. The Optum algorithm takes in years of patient data and gives each patient a risk score that is used to determine enrollment in a high-risk patient care program ("Algorithms & Population Health Management", n.d.). My STS research centers on an audit of the Optum algorithm done by Obermeyer et al that found that Black patients of equivalent health were less likely to be enrolled in the program compared to white patients (Obermeyer, 2019). My claim is that by using Actor-Network Theory (ANT) the adversarial actor that introduces systemic racism into the Optum algorithm network can be determined to be the label. I demonstrate that the choice of past healthcare costs as the label introduces the racial bias by considering simulated algorithms with different labels that do not contain a health gap between enrolled white and Black patients. I also describe aspects of our healthcare system that lead to a racial gap between healthcare costs and health needs. The goal of this analysis is to evoke deeper thought in engineers about how their design choices can imbue systemic inequalities and propose ANT as a way to understand what social or conceptual factors might influence technical elements.
My technical work gave me a better understanding of the technical reasons for design choices that go into algorithms. Working on the STS research paper, made evident that design choices also have social impacts. In the field of computer science technologies, we can erroneously believe that the computer is immune to the biases and flaws that plague humans, but studying the Optum algorithm showed me that technical actors have many social and conceptual influences on them. Doing these projects concurrently prompted a greater reflection on the social world that could be created by the ARK device. The most constant way this manifested was asking ourselves whether we were serving all patient populations.

References
Algorithms & population health management. (n.d.). Optum. Retrieved November 1, 2020, from https://www.optum.com/business/health-insights/algorithms-human-touch.html
Obermeyer, Z., Powers, B., Vogeli, C., & Mullainathan, S. (2019). Dissecting racial bias in an algorithm used to manage the health of populations. Science, 366(6464), 447–453. https://doi.org/10.1126/science.aax2342
Romer, L. M., & Polkey, M. I. (2008). Exercise-induced respiratory muscle fatigue: Implications for performance. Journal of Applied Physiology, 104(3), 879–888. https://doi.org/10.1152/japplphysiol.01157.2007
Signs of Respiratory Distress. (n.d.). John Hopkins Medicine. Retrieved October 25, 2020, from https://www.hopkinsmedicine.org/health/conditions-and-diseases/signs-of-respiratory-distress

School of Engineering and Applied Science

Bachelor of Science in Biomedical Engineering

Technical Advisor: Shrirang Gadrey MBBS MPH

STS Advisor: Benjamin Laugelli PhD

Technical Team Members: Sarah Innis, Julia Shanno