A machine-learning based analysis for the recognition of progressive central hypovolemia

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

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.

Support Vector Machine Based Monitoring of Cardio-Cerebrovascular Reserve during Simulated Hemorrhage

Introduction: In the initial phase of hypovolemic shock, mean blood pressure (BP) is maintained by sympathetically mediated vasoconstriction rendering BP monitoring insensitive to detect blood loss early. Late detection can result in reduced tissue oxygenation and eventually cellular death. We hypothesized that a machine learning algorithm that interprets currently used and new hemodynamic parameters could facilitate in the detection of impending hypovolemic shock.

Method: In 42 (27 female) young [mean (sd): 24 (4) years], healthy subjects central blood volume (CBV) was progressively reduced by application of −50 mmHg lower body negative pressure until the onset of pre-syncope. A support vector machine was trained to classify samples into normovolemia (class 0), initial phase of CBV reduction (class 1) or advanced CBV reduction (class 2). Nine models making use of different features were computed to compare sensitivity and specificity of different non-invasive hemodynamic derived signals. Model features included: volumetric hemodynamic parameters (stroke volume and cardiac output), BP curve dynamics, near-infrared spectroscopy determined cortical brain oxygenation, end-tidal carbon dioxide pressure, thoracic bio-impedance, and middle cerebral artery transcranial Doppler (TCD) blood flow velocity. Model performance was tested by quantifying the predictions with three methods: sensitivity and specificity, absolute error, and quantification of the log odds ratio of class 2 vs. class 0 probability estimates.

Results: The combination with maximal sensitivity and specificity for classes 1 and 2 was found for the model comprising volumetric features (class 1: 0.73–0.98 and class 2: 0.56–0.96). Overall lowest model error was found for the models comprising TCD curve hemodynamics. Using probability estimates the best combination of sensitivity for class 1 (0.67) and specificity (0.87) was found for the model that contained the TCD cerebral blood flow velocity derived pulse height. The highest combination for class 2 was found for the model with the volumetric features (0.72 and 0.91).

Conclusion: The most sensitive models for the detection of advanced CBV reduction comprised data that describe features from volumetric parameters and from cerebral blood flow velocity hemodynamics. In a validated model of hemorrhage in humans these parameters provide the best indication of the progression of central hypovolemia.