Bleeding Or Active? Validation Of A Machine-learning Algorithm For Remote Determination Of Blood Volume Status

Abstract

Due to the lack of remote monitoring capabilities, combat medics cannot currently distinguish a wounded and bleeding soldier from a soldier who is engaged in combat unless they have physical access to them. We have previously shown that traditionally measured vital signs may not provide the specificity required to distinguish between a bleeding or physically active soldier, but a measure of volume status (e.g., pulse pressure (PP)), may be able to achieve this purpose. PURPOSE: To assess whether a machine-learning algorithm programmed within a wearable, non-invasive armband can track changes in PP and therefore distinguish between the conditions of central hypovolemia and exercise. METHODS: Twenty five human subjects underwent progressive central hypovolemia via lower body negative pressure (LBNP); 8 of these subjects also participated in a supine cycle exercise protocol (≥5 weeks later). Exercise workloads were determined by matching heart rate (HR) responses at each LBNP level. HR (via ECG) and arterial blood pressure (via Finometer®) were measured continuously. Subjects wore a SenseWear® Pro2 Armband that measured heat flux, HR, skin temperature, galvanic skin response, and 3-axis acceleration. These measurements were used to develop a machine-learning algorithm to predict changes in PP as an index of central blood volume under both conditions; these predictions were retrospectively compared against actual PP values obtained from the Finometer®. An additional model was created to determine if the accuracy, sensitivity, and specificity of unknown data points could be correctly classified into these two conditions (N = 8). RESULTS: Predicted versus actual PP values were highly correlated for both LBNP (r = 0.84; n = 25) and exercise (r = 0.86; n = 8). The machine learning algorithm was able to distinguish between LBNP and exercise with high accuracy (94%), sensitivity (95%), and specificity (92%). CONCLUSIONS: These results demonstrate that a wearable armband with a machine-learning algorithm was able to reliably track changes in PP during both central hypovolemia and exercise. This device could provide remote triage capabilities for future battlefield use.