Detecting central hypovolemia in simulated hypovolemic shock by automated feature extraction with principal component analysis

Abstract

Assessment of the volume status by blood pressure ( BP ) monitoring is difficult, since baroreflex control of BP makes it insensitive to blood loss up to about one liter. We hypothesized that a machine learning model recognizes the progression of central hypovolemia toward presyncope by extracting information of the noninvasive blood pressure waveform parametrized through principal component analysis. This was tested in healthy volunteers exposed to simulated hemorrhage by lower body negative pressure ( LBNP ). Fifty‐six healthy volunteers were subjected to progressive central hypovolemia. A support vector machine was trained on the blood pressure waveform. Three classes of progressive stages of hypovolemia were defined. The model was optimized for the number of principal components and regularization parameter for penalizing misclassification (cost): C. Model performance was expressed as accuracy, mean squared error ( MSE ), and kappa statistic (inter‐rater agreement). Forty‐six subjects developed presyncope of which 41 showed an increase in model classification severity from baseline to presyncope. In five of the remaining nine subjects (1 was excluded) it stagnated. Classification of samples during baseline and end‐stage LBNP had the highest accuracy (95% and 50%, respectively). Baseline and first stage of LBNP demonstrated the lowest MSE (0.01 respectively 0.32). Model MSE and accuracy did not improve for C values exceeding 0.01. Adding more than five principal components did not further improve accuracy or MSE . Increment in kappa halted after 10 principal components had been added. Automated feature extraction of the blood pressure waveform allows modeling of progressive hypovolemia with a support vector machine. The model distinguishes classes between baseline and presyncope.

Wearable Sensors Incorporating Compensatory Reserve Measurement for Advancing Physiological Monitoring in Critically Injured Trauma Patients

“…When this capacity to compensate becomes depleted, a state of decompensated shock occurs. Clinically, a compensatory reserve measurement (CRM) can be obtained from assessment of changing arterial pressure waveform morphology associated with changes in compensation [ 14 , 18 , 21 , 23 , 30 , 38 , 39 , 40 , 41 , 42 , 43 , 44 , 45 , 46 , 47 , 48 , 49 , 50 , 51 , 52 , 53 , 54 ]. Figure 2 illustrates that each arterial waveform consists of two primary waves: (1) an ‘ejected’ wave with features that are dictated by all compensatory mechanisms that influence myocardial function; and (2) a ‘reflective’ wave with features that are influenced by all compensatory mechanisms involved in the control of peripheral blood flow [ 14 , 22 ].…”

“…Optimal management of significant traumatic hemorrhage and other compromising clinical conditions is often delayed by failure to recognize a medical crisis due to the current reliance on traditional vital signs and/or other standard physiological measures that represent a limited assessment of a totally integrated compensatory response [ 22 , 24 , 25 , 26 , 27 , 28 , 29 , 54 , 61 , 64 ]. In this regard, the value of monitoring the arterial waveform morphology for early detection of a clinical crisis using a CRM algorithm has been well documented during actual controlled human hemorrhage in the laboratory setting [ 14 , 22 , 25 , 38 , 39 , 40 , 41 , 42 , 44 , 50 , 52 , 53 , 61 , 64 , 65 ], and translated to early recognition of hypovolemia and hypotension when used by first responders during simulated emergencies training exercises [ 66 , 67 ], and in hospital critical care settings [ 20 , 21 , 43 , 45 , 46 , 47 , 49 , 51 , 60 , 68 , 69 , 70 , 71 , 72 ]. The comparative data regarding sensitivity, specificity and diagnostic accuracy of various monitoring technologies presented in this review provide compelling support for the notion that the development of wearable sensors must include an ability to capture analog signals that allow for continuous real-time analysis of changes in ...…”

Central Hypovolemia Detection During Environmental Stress—A Role for Artificial Intelligence?

“…Traditional patient monitoring in the emergency ward and the operating room includes HR, BP, electrocardiogram, and peripheral oxygen saturation but their use as predictors for incipient central hypovolemia is rather limited. Baroreflex control of BP makes it insensitive to blood loss up to about one liter, rendering assessment of volume status by BP monitoring not possible (Harms et al, 2003 ; van der Ster et al, 2018b ). With the progression of central hypovolemia, cardiac preload declines until the tipping point where it has become too low to maintain a sufficient CO and when the limits of vasomotor reserve available for vasoconstriction have been reached BP drops (Schondorf and Wieling, 2000 ; Fu et al, 2004 ; Schiller et al, 2017 ).…”

“…In a laboratory model of hemorrhage simulated by progressive central hypovolemia in healthy subjects by submitting them to either LBNP, HUT or both the global hypothesis tested was that AI-based methodologies may assist in monitoring and accordingly predict the progression from normo- to hypovolemia toward presyncope/cardiovascular collapse by extracting information of biomedical signals otherwise not routinely available. Figure 2 , panel B, summarizes modeling of the non-invasive BP waveform (van der Ster et al, 2018b , 2020 ). During simulated hemorrhage in healthy subjects, volumetric parameters together with CBF velocity hemodynamics provided the most sensitive indication of the progression of central hypovolemia ( Figure 2 , panel C) (van der Ster et al, 2018a ; van der Ster, 2019 ).…”

“…BP, blood pressure; PetCO2, end-tidal carbon dioxide partial pressure; NIRS, near infrared spectroscopy; TCD, transcranial Doppler; MCAv, middle cerebral artery blood flow velocity; LBNP, lower body “negative” or sub-atmospheric pressure. Modified from van der Ster et al (2018a , b) .…”