Trauma patients with "compensated" internal hemorrhage may not be identified with standard medical monitors until signs of shock appear, at which point it may be difficult or too late to pursue life-saving interventions. We tested the hypothesis that a novel machine-learning model called the compensatory reserve index (CRI) could differentiate tolerance to acute volume loss of individuals well in advance of changes in stroke volume (SV) or standard vital signs. Two hundred one healthy humans underwent progressive lower body negative pressure (LBNP) until the onset of hemodynamic instability (decompensation). Continuously measured photoplethysmogram signals were used to estimate SV and develop a model for estimating CRI. Validation of the CRI was tested on 101 subjects who were classified into two groups: low tolerance (LT; n = 33) and high tolerance (HT; n = 68) to LBNP (mean LBNP time: LT = 16.23 min vs. HT = 25.86 min). On an arbitrary scale of 1 to 0, the LT group CRI reached 0.6 at an average time of 5.27 ± 1.18 (95% confidence interval) min followed by 0.3 at 11.39 ± 1.14 min. In comparison, the HT group reached CRI of 0.6 at 7.62 ± 0.94 min followed by 0.3 at 15.35 ± 1.03 min. Changes in heart rate, blood pressure, and SV did not differentiate HT from LT groups. Machine modeling of the photoplethysmogram response to reduced central blood volume can accurately trend individual-specific progression to hemodynamic decompensation. These findings foretell early identification of blood loss, anticipating hemodynamic instability, and timely application of life-saving interventions.
Keywords: blood pressure; hemorrhage; modeling; shock; stroke volume.