Tamara Lambert, MPH, M.Eng
BME PhD Defense Presentation

Date: 2024-06-19
Time: 3:00 PM -5:00 PM
Location / Meeting Link: TSRB 523a / https://emory.zoom.us/j/94283721482?pwd=N1RQdGJSMlFHakJydjJaN3V1Qk5FQT09 

 


Committee Members:
Omer T. Inan, Ph.D (Advisor); Eva L. Dyer, Ph.D; Rishikesan Kamaleswaran, Ph.D; Jacob Kimball, Ph.D; Ying Zhang, Ph.D


Title: Towards real-time estimation of decompensation status in acute pathological conditions leveraging wearable technology and machine learning

Abstract:
Wearable technologies are rapidly gaining traction and utility in the healthcare space. These technologies provide non-invasive, real-time monitoring of parameters indicative of the patient's health status and can thus detect potential abnormalities. Although advances in wearable technology have demonstrated robust clinical utility such as monitoring vital signs, there are other areas where the clinical use of wearable technologies remain underexplored. To this end, this work seeks to elucidate the use of wearable technology in predicting the presence of decompensated states in two clinical applications, opioid use disorder (OUD) and hypovolemia. Current approaches to detecting these pathological conditions have several limitations. The diagnosis of withdrawal from opioids relies on the Clinical Opiate Withdrawal Scale (COWS) and clinician observation, which can be subjective and prone to bias. Current methods of estimating the blood volume decompensation status (BVDS) in hypovolemic patients are either invasive, require knowledge of the patient’s baseline BVDS, or do not provide accurate and timely indication of decompensation due to the body’s compensatory mechanisms. These limitations can lead to a clinician missing a patient at elevated risk of relapse and overdose in the case of OUD, and emergency personnel unable to prioritize casualties, which is critical in wartime or a mass-casualty event in the case of BVDS. In Aim 1, this work demonstrates a novel methodology using a tri-axial accelerometer to detect the presence of worsening opioid withdrawal symptoms. In Aim 2, we compare several normalization techniques for individual baseline-free estimation of hypovolemic status, demonstrating the ability of our model to achieve accurate results in the absence of subject specific baseline data. Finally, in Aim 3, we demonstrate the use of non-invasive cardiac signals to estimate changes in blood biomarker levels. The results of this work will contribute to new paradigms using wearable sensing technology to detect decompensation in a variety of pathological conditions.