Unifying the Estimation of Blood Volume Decompensation Status in a Porcine Model of Relative and Absolute Hypovolemia Via Wearable Sensing

Classification of Blood Volume Decompensation State via Machine Learning Analysis of Multi-Modal Wearable-Compatible Physiological Signals

“…On the one hand, the relevance of PEP/LVET metrics may be supported by the physics of cardiac function, which suggests that LVET may decrease with absolute hypovolemia due to the reduction in the cardiac stroke volume (SV) [ 40 , 41 ]. Hence, PEP/LVET increases as cardiac SV decreases, suggesting that an increase in PEP/LVET can be an indicator of absolute hypovolemia (in fact, we experimentally demonstrated the value of PEP/LVET in quantifying absolute hypovolemia in our prior work [ 24 , 25 ]). From this standpoint, the positive regression coefficients between absolute hypovolemia and PEP AO /LVET and PEP H /LVET are plausible.…”

“…Experiments were performed in six Yorkshire swine (age: 114–150 days, weight: 52–71 kg) under the approval of the IACUC at the Georgia Institute of Technology (A100276), Translational Testing and Training Labs, Inc. (GT48P), and the Department of Navy Bureau of Medicine and Surgery. Full details of the experimental protocol are described in our prior work [ 24 , 25 ]. In brief, each animal was anesthetized and subject to a baseline period.…”

Wearable Sensors and Machine Learning for Hypovolemia Problems in Occupational, Military and Sports Medicine: Physiological Basis, Hardware and Algorithms

“…Encouraged by the results from CRM, a collaboration led by Inan and colleagues recently developed the blood volume decompensation status (BVDS) metric [12,58]. The goal of the BVDS metric builds from that of the CRM-to develop a single metric that represents the integrative compensatory response based on some aspect of PPG feature changes, and thus can be used to represent an individual's compensatory reserve or decompensation status.…”

“…An initial model was developed using only the hemorrhage data recorded from the noninvasive sensors and compared to another model created with an analogous feature set extracted from simultaneously acquired invasive catheter blood pressure waveforms [58]. The BVDS model was further developed as data from the relative and absolute portions of the experiment were used together to train the random forest regression model with leave-one-subject-out cross validation to create a more generic metric of decompensation status [12]. The feature importance output by this model is shown in Figure 12, indicating that electromechanical features of cardiac performance were the most important predictors.…”

“…However, measures of standard vital signs are limited in their ability to provide individual-specific information about those at greatest health risk or with performance impairments because they are not based on the understanding of underlying integrated physiological mechanisms associated with the adverse outcomes from hypovolemia. Compensatory reserve and decompensation measures demonstrate great clinical promise for monitoring and treating hemorrhage hypovolemia [7,12], and we argue these approaches can be effectively applied to a broader set of hypovolemia problems [13]. However, the broader use of compensatory reserve or decompensation measures, particularly in austere field conditions where multiple stressors are combined with hypovolemia, pose several technology and algorithmic challenges that preclude current approaches from translating successfully to field settings.…”