Real-Time Comparison of Machine Learning-Enabled Devices for Measuring Compensatory Reserve Status.

In this work, the real-time measurement of compensatory reserve metrics is compared during a simulated central hypovolemia clinical research study. We hypothesized that values generated from the compensatory reserve measurement (CRM) device would be statistically similar to those values generated from the compensatory reserve index (CRI) device. A study was conducted to collect data from noninvasive photoplethysmography sensors to validate the CRM against an FDA-cleared CRI device. Participants were placed in a lower body negative pressure (LBNP) chamber and underwent exposure to a stepwise protocol of increasing negative pressure. Data from both algorithmic devices were evaluated using multiple approaches: median error, median absolute error, and pooled Pearson correlations. Data were compared from 20 participants through the duration of the experiment. Pooled correlation analysis between CRM and CRI for all subjects and LBNP steps achieved R2 = 0.859, showing strong linear agreement. Bland-Altman analysis revealed a near-zero mean bias between CRM and CRI (0.13%). Each metric allowed for a mean early measurement time of 20.52 min for CRM and 17.92 min for CRI. Consistent with our hypothesis, the CRM device performed similarly to CRI, demonstrating that both models can track compensatory status changes with progressive central hypovolemia. The early, real-time predictive tracking of compensatory changes can be valuable for continuous monitoring and triaging in prehospital trauma environments.