Evaluation of a novel closed-loop fluid-administration system based on dynamic predictors of fluid responsiveness: an in silico simulation study

Abstract

Introduction: Dynamic predictors of fluid responsiveness have made automated management of fluid resuscitation more practical. We present initial simulation data for a novel closed-loop fluid-management algorithm (LIR, Learning Intravenous Resuscitator).

Methods: The performance of the closed-loop algorithm was tested in three phases by using a patient simulator including a pulse-pressure variation output. In the first phase, LIR was tested in three different hemorrhage scenarios and compared with no management. In the second phase, we compared LIR with 20 practicing anesthesiologists for the management of a simulated hemorrhage scenario. In the third phase, LIR was tested under conditions of noise and artifact in the dynamic predictor.

Results: In the first phase, we observed a significant difference between the unmanaged and the LIR groups in moderate to large hemorrhages in heart rate (76 ± 8 versus 141 ± 29 beats/min), mean arterial pressure (91 ± 6 versus 59 ± 26 mm Hg), and cardiac output (CO; (6.4 ± 0.9 versus 3.2 ± 1.8 L/min) (P < 0.005 for all comparisons). In the second phase, LIR intervened significantly earlier than the practitioners (16.0 ± 1.3 minutes versus 21.5 ± 5.6 minutes; P < 0.05) and gave more total fluid (2,675 ± 244 ml versus 1,968 ± 644 ml; P < 0.05). The mean CO was higher in the LIR group than in the practitioner group (5.9 ± 0.2 versus 5.2 ± 0.6 L/min; P < 0.05). Finally, in the third phase, despite the addition of noise to the pulse-pressure variation value, no significant difference was found across conditions in mean, final, or minimum CO.

Conclusion: These data demonstrate that LIR is an effective volumetric resuscitator in simulated hemorrhage scenarios and improved physician management of the simulated hemorrhages.

Figures

Figure 1

Rule-based component of the controller algorithm. The controller uses patient hemodynamic parameters (primarily pulse-pressure variation, but also cardiac output, mean arterial pressure, and heart rate) that are compared with the dataset and a probability of positive response assigned based on the population data. This probability and the hemodynamic data are then fed into the rule-based component of the controller. CO, Cardiac output.

Figure 2

Cardiac output in Phase 2 groups; closed-loop system versus practitioner management during a simulated hemorrhage scenario. Each line represents a single case. Once the hemorrhage began, the LIR-managed groups intervened significantly earlier than the practitioner group and gave more total fluid. The mean, minimum, and final cardiac output was higher in both LIR-managed groups than in the practitioner group, and the coefficient of variance was lower. LIR, Learning Intravenous Resuscitator.

Figure 3

Mean arterial pressure in Phase 2 groups: closed-loop system versus practitioner management during a simulated hemorrhage scenario. Each line represents a single case. We observed no difference in mean arterial pressure between intervention groups, but all were significantly higher than those in the unmanaged group.