Keywords

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

General Principles of Fluid Therapy and Hemodynamic Optimization

Before a closed-loop system can be designed for any given task, there must be a clearly defined control objective. While this may seem obvious, in practice it can be a significant challenge. In the case of IV fluid administration, for example, there is no medical consensus on what constitutes optimal fluid therapy. It’s clear that hypovolemia is undesirable and increases morbidity and mortality, and it’s also clear that hypervolemia leads to complications as well. What constitutes “appropriate” resuscitation is not as clear. The first requirement for closed-loop fluid administration, therefore, is to define an endpoint of resuscitation.

The goal of IV fluid administration is to expand intravascular volume and in so doing improve blood flow to the tissues. Adequate tissue perfusion depends on both cardiac output and mean arterial pressure; the former ensures adequate oxygen and nutrient delivery and removal of waste while the latter ensures adequate driving pressure to force blood into capillaries and balance flow among the different tissues of the body which will have different needs and pressure requirements depending on the circumstances (Fig. 17.1). Fluid administration can increase cardiac output by increasing venous return. The increase in cardiac output may, in some cases, lead to an increase in mean arterial pressure as well, but this is not always the case. Because blood pressure arises from both cardiac output and systemic vascular resistance, the relationship between cardiac output and arterial pressure is not fixed. Despite this, standard practice in many clinical settings focuses fluid administration on arterial pressure alone, and flow monitoring (in the form of a cardiac output monitor of some kind) is overlooked.

Fig. 17.1
figure 1

Basic hemodynamic interrelationships. No element of hemodynamics can really be considered in isolation due to the interconnectedness of all of the components. Blood pressure, for example, arises from both vascular tone and cardiac output. Cardiac output, in turn, is dependent on both stroke volume and heart rate. To change any one of these components would alter the others as well. Thus, hemodynamic function must really be considered from a system standpoint as consideration of only individual elements will not provide a complete picture

Full size image

The recent focus on cardiac output optimization/maximization in moderate- to high-risk surgery, however, has begun to change this trend in favor of administering fluid based on cardiac output endpoints instead of mean arterial pressure alone or generalized formulas. In this approach, fluid is administered until cardiac stroke volume no longer increases by some predetermined percentage in response to the bolus (e.g., a 10 % increase in stroke volume after 200 mL of fluid) [1]. Other approaches include dynamic predictors of fluid responsiveness like pulse pressure variation and stroke volume variation to reduce the number of unneeded fluid boluses. In appropriate patients (mechanically ventilated with 8 mL/kg tidal volumes, without arrhythmias, and without significant valvular or right ventricular pathophysiology) [2], these parameters can be used to predict whether the patient will increase stroke volume in response to fluid administration [3]. Fluid administration protocols based on minimization of stroke volume variation, for example, have shown postoperative benefits [4], as have those based on pulse pressure variation [5] and plethysmograph variability [6].

Given the growing evidence supporting these approaches and the improved outcomes they are associated with, stroke volume and cardiac output maximization or optimization seems a natural target endpoint for closed-loop fluid administration. There is a subtle but important difference between the terms “optimization” and “maximization.” Strictly speaking, stroke volume or cardiac output “maximization” would refer to the administration of IV fluid to drive the patient to the very top of the plateau phase of the Frank-Starling relationship (Fig. 17.2). While this would maximize oxygen delivery and blood flow, it may also lead to over-resuscitation because of the preload volumes required to achieve this result, especially in cardiovascularly fit patients who have high cardiac reserve [7]. “Optimization” would ideally mean matching cardiac output and oxygen delivery to oxygen consumption. At present, however, there is no convenient or reliable method to determine global oxygen consumption, nor would a global level guarantee that all individual tissues were adequately perfused because of regional differences in metabolic rate and blood flow. Thus, we more often use the term “optimization” to refer to the administration of fluid to bring the patient near the plateau phase of the Frank-Starling relationship, balancing the amount of fluid required to make further gains with the metabolic needs of the patient.

Fig. 17.2
figure 2

Starling curve and cardiac output optimization. Because of the nonlinear response of the left ventricle to preload, the beneficial effects of fluid administration on stroke volume and cardiac output are subject to the law of diminishing returns. In the figure, it is easy to see that for a given fixed increase in stroke volume, increasingly larger preloads are required until at some point no further increase is even possible (the plateau phase). “Maximization” of cardiac output would require a relatively large volume of fluid to be given to drive the stroke volume to this maximal point. On the other hand, “optimization” of stroke volume could be considered achieved by merely approaching the plateau phase of the Starling curve. This could achieve 80–90 % of the maximal stroke volume with substantially less fluid loading required

Full size image

The best evidence we have at present, then, is that in moderate- to high-risk patients the goal of fluid administration should be cardiac output optimization. That means increasing the stroke volume and cardiac output to near (but not necessarily on) the plateau of the Starling curve, using as feedback either percentage change stroke volume/cardiac output, minimization of a dynamic predictor like pulse pressure variation or stroke volume variation, or some combination of both.

Feedback Parameter Selection (and Historical Closed Loops)

Table 17.1 reviews some of the parameters that are relevant for fluid administration and might be used in a closed-loop resuscitation system. These parameters are discussed below along with studies that have been done using these parameters to guide automated resuscitation. Published studies on closed-loop fluid management are summarized in Table 17.2.

Table 17.1 Potential feedback parameters for use in a closed-loop fluid delivery system
Full size table
Table 17.2 Closed-loop fluid management studies
Full size table

Urine Output

The very first closed-loop fluid administration system was developed in Utah in the early 1980s and was based on urine output [8]. This system was actually used clinically in intensive care for a time but was subsequently phased out of use when one of the developers left the university. Recent work in sheep using a proportional-integral-derivative (PID) controller to resuscitate after burn injuries showed more stable urine outputs than did manual adjustments [9]. This is typical of closed-loop control: frequently, there is no large difference in mean values between closed-loop control and manual control, but the closed-loop shows a much narrower range of deviations from the mean.

One challenge with urine output as a feedback parameter is that it is a late sign of hypovolemia, is relatively nonspecific, and is strongly affected by pharmacologic agents and blood pressure. In ICU conditions, it may be a reasonable target parameter for closed-loop fluid administration if the goal is to maintain net fluid balance. In more dynamic environments like the operating room or trauma settings, however, urine output would be of limited value as a feedback parameter.

Heart Rate

As a primary determinant of cardiac output and oxygen delivery, heart rate is intrinsically tied to volume status and resuscitation. Heart rate is obviously a very general guide to volume status, but there is little doubt that tachycardia is an expected response to hypovolemia. Like urine output, however, tachycardia is a late sign of hypovolemia and generally only appears when volume deficit is severe (15 % or more of total blood volume). Additionally, normocardia is not an assurance of adequate blood volume, especially in the presence of beta-blockade or cardiac pathology. Moreover, heart rate is normally dynamic in healthy patients and is affected by pain, narcotics, sympathomimetics, and more. Perhaps because of this complexity, heart rate has never been used in closed-loop resuscitation as a control parameter.

The real utility and necessity of heart rate monitoring in closed-loop resuscitation is that heart rate is intrinsically tied to all other hemodynamic parameters and must be accounted for in any comprehensive resuscitation algorithm. For example, a computer algorithm based on cardiac output alone would not be able to differentiate between a stroke volume of 80 with a heart rate of 60 (CO = 4.8) and a stroke volume of 40 with a heart rate of 120 (CO = 4.8), even though these are obviously very different physiologic states in an adult. Note that using stroke volume instead of cardiac output does not entirely circumvent this problem, as heart rate still affects cardiac filling time and therefore influences stroke volume. Thus, heart rate is likely a required consideration in closed-loop fluid therapy and hemodynamic optimization. Additionally, changes in heart rate may provide valuable information about sympathetic outflow and pharmacologic interventions.

Mean Arterial Pressure

Mean arterial pressure is an easily measureable parameter and can be measured continuously, and now even noninvasive monitors are providing continuous waveform data. Since arterial pressure is required to drive blood into end-organ capillaries, a closed-loop hemodynamic controller would almost certainly want to include a continuous blood pressure measurement in the algorithm.

Anesthesiologists have traditionally focused resuscitation efforts towards management of blood pressure, and recent survey data has shown that this focus persists today. Blood pressure alone, however, is a poor predictor of preload dependency, and blood volume must be significantly reduced before arterial pressure begins to fall. Likewise, hypertension is not a specific indication of volume overload. Thus, as with heart rate, mean arterial pressure is unlikely to be useful as a lone endpoint but rather as a single parameter in a multi-input system.

Central Venous Pressure and Wedge Pressure

Central venous pressure and pulmonary capillary wedge pressure have historically been considered the best guides to resuscitation in intensive care and high-risk patients. Despite the common belief that these values accurately reflect left ventricular preload, however, these parameters are very inaccurate predictors of fluid responsiveness [10–12]. Even so, they have been used successfully to guide closed-loop resuscitation for postoperative autotransfusion [13] and for volume replacement during continuous hemofiltration [14]. It is conceivable that in otherwise stable patients who are not being subjected to changing pharmacologic therapies, surgical stimulation, anesthetic levels, or other dynamic forces, that these static parameters may generally trend with volume status and be adequate for guidance of resuscitation. Generally speaking, however, these static parameters are unreliable in most clinical conditions and are probably of limited value without other information.

Cardiac Output and Stroke Volume

Cardiac output and stroke volume are obvious choices for closed-loop control, primarily because they are principle determinants of oxygen delivery and have outcomes studies to support their use [15]. As noted above, cardiac output optimization cannot be performed without some consideration of heart rate. However, while increasing heart rate to improve cardiac output is a potentially useful strategy, it should be considered only when blood volume is already optimized, and tachycardia may be detrimental in patients with ischemic cardiac disease. Thus, most of the time when we speak of cardiac output optimization, we mean stroke volume optimization.

Either way, cardiac output and stroke volume are natural targets for closed-loop control, especially if the controller can account for the other components of hemodynamics (heart rate, arterial pressure, and vascular tone). The controller need not actually control these parameters as well, but an ability to utilize the additional information could add sophistication to the control algorithm. Stroke volume has been used as a primary control parameter in simulation studies on closed-loop control [16].

Dynamic Predictors

The dynamic predictors of fluid responsiveness – measurable parameters that arise from the interaction of the cardiovascular system and thorax during positive-pressure ventilation [17] – are the most accurate predictors of volume responsiveness yet described [18]. As such, they are invaluable for closed-loop resuscitation. The limiting factor for these parameters is that they are only accurate under specific conditions: tidal volume at least 8 mL/kg and positive-pressure ventilation, no arrhythmias, no valvular disease, closed-chest conditions, and normal right ventricular function. Furthermore, conditions like abdominal insufflation, prone positioning, and more have not been thoroughly studied and may reduce or invalidate the utility of these parameters. Any closed-loop designed for hemodynamic optimization should certainly be capable of using these predictors when the conditions are appropriate, but should also be able to function in their absence if it is not to be subject to the same set of limitations.

Mixed Venous Oxygen Saturation

Measurement of mixed venous oxygen provides a very sensitive indication of the balance between oxygen delivery and oxygen consumption in the body as a whole. Low values result from inadequate delivery and increased uptake in the periphery, while high values result from shunting and inadequate uptake. Alone this information would not indicate the cause of inadequate delivery (e.g., low cardiac output from hypovolemia versus heart failure) but if combined with cardiac output and other data could provide a very complete overall picture of hemodynamics. The challenge with mixed venous oxygen is that accurate measurement can be difficult due to differences in venous return from different parts of the body, and even if mixed venous oxygen is normal, it does not guarantee all tissues are adequately perfused. In septic shock supported with vasopressors, for example, global mixed oxygen may be in the normal range in the presence of some peripheral shunting (which would increase the value) combined with bowel ischemia (which would decrease the value).

Tissue SpO2

Tissue oxygen levels are also closely related to cardiac output and have been used to guide closed-loop fluid resuscitation [19]. Of course, tissue oxygen levels will also be affected by hemoglobin concentration, arterial PaO2, and tissue metabolism. Additionally, due to differences in regional blood flow, tissue oxygen levels in some tissues (muscle) may be adequate while other tissues (bowel) become ischemic, especially in the presence of sympathomimetic agents which cause redistributions in blood flow. Even given these limitations, ultimately our goal with resuscitation is to restore oxygen delivery and blood flow to capillary beds, so there is a strong intuitive rationale to incorporate tissue oxygenation in some form into a closed-loop control scheme for resuscitation if simultaneous measurement in many body tissues becomes feasible.

Specific Monitoring Considerations

In addition to the specific parameter or parameters used to guide fluid administration and resuscitation, some consideration must be given to the monitoring system used to provide that information. Some parameters, like heart rate and mean arterial pressure, can be monitored very accurately from a variety of sources. Other parameters, like stroke volume and tissue SpO2, may be dependent on the method and location of measurement. For example, disagreement between values obtained from esophageal Doppler measurements and thermodilution measurement techniques is up to 20 % [20], and larger disagreement has been found between thermodilution and measurements of cardiac output derived from arterial waveform analysis [21], especially in patients with disturbances in vascular tone.

Disagreements between measurement approaches do not mean that these measurements are not still potentially useful for guidance of resuscitation or closed-loop control. Noninvasive and minimally invasive systems can still provide good information with reduced risk to patients as compared with more invasive monitoring. We must be aware of their limitations, however, and include those limitations in our clinical decision-making. Thus, use of these systems in closed-loop control will require control schemes that include the possibility of error in measurement and protect the patient from adverse events as a result.

Certainly monitor accuracy will improve with time and at some point error will no longer be as large an issue. Until that time, however, the trade-offs between clinical utility and limitations of current monitors must be a design consideration in any closed-loop for hemodynamic management.

Controller Types

Based on the feedback parameters available for guidance of closed-loop resuscitation discussed above, it seems obvious that rather than an individual parameter, a multiple-parameter strategy makes the most sense for control. Figure 17.3 shows an example of why multiparameter feedback may be not just advisable but required for closed-loop fluid administration. Figure 17.3a shows a sample of cardiac stroke volume as recorded during an animal hemorrhage study [22]. Looking at only the stroke volume, it is very difficult to determine the causes of the changes seen at point 1 and point 2. For example, the increase at point 1 could result from a fluid bolus, sympathetic activity, or even a sudden fall in left ventricular afterload. Likewise, the drop in stroke volume at point 2 could result from hemorrhage, cardiac ischemia, or a sudden rise in afterload.

Fig. 17.3
figure 3

Sample data from animal studies showing single versus multiparameter feedback. Hemodynamic data from unpublished animal hemorrhage studies. The top graph (a) shows only stroke volume (SV). The bottom graph (b) shows heart rate (HR), mean arterial pressure (MAP), and systemic vascular resistance (SVR) as well. It is much easier to determine what occurred at points 1 and 2 with the additional data in the second graph (see text for details)

Full size image

Figure 17.3b shows the same data, this time including heart rate, mean arterial pressure, and systemic vascular resistance. It becomes easier to determine the causes of the changes seen in stroke volume with this new data. At point 1, we now see that stroke volume rose along with arterial pressure, vascular resistance, and heart rate. This is consistent with sympathetic activity or a sympathomimetic, in this case a dose of ephedrine. At point 2, we now see that stroke volume and heart rate fell while vascular tone and arterial pressure rose dramatically. This is consistent with an alpha-agonist effect, in this case a dose of phenylephrine. Because of the interconnected nature of hemodynamics, it is only with the additional information provided from the other parameters that we can make inferences about observed changes in stroke volume.

There are a variety of control schemes which handle multiple-parameter input and output. A primary controlled variable could be chosen (e.g., cardiac output), with a single available intervention (perhaps fluid administration), with other monitored parameters used to inform decision-making about how to achieve control at any given time. Alternately, a true multi-input, multi-output (MIMO) controller could be designed to monitor multiple parameters simultaneously using multiple interventions (like fluid and pharmacologic agents).

Proportional-integral-derivative control (PID) is one of the most commonly used control schemes in control engineering. The three terms of the controller refer to the present, past, and future error between a target value for some measured parameter and the measured value from the system. PID controllers require tuning of the weights of the components to prevent oscillations in the system but once developed can provide very stable and robust control. One limitation of this control type for closed-loop hemodynamic applications is that PID control is best suited for single-parameter measurement. Additionally, for applications like fluid administration, the nonlinear and dynamic response of hemodynamics to volume expansion makes PID control less effective. It would be well suited, however, for control of a pharmacologic agent like phenylephrine to maintain a target blood pressure.

Model-based controllers use an underlying model of the system to be controlled to make predictions about the response to interventions, and these predictions are then used to guide control. Controllers like these are made more robust by incorporating correction factors for differences or error between the internal model and the actual system under control (referred to as “plant-model mismatch”). Model-based controllers, of course, require a reasonably predictive model of the system to be effective but are appropriate for MIMO applications if sufficient testing and fine tuning of the controller are done.

Artificial neural network-based controllers are very well suited to MIMO applications. These controllers are designed to mimic a small network of neurons, with inputs being “processed” through one or more layer of nodes before outputs are generated. Neural networks range from simple, single-layer, single-direction designs to multilayer designs where later layers can feedback on earlier layers or even themselves. The chief challenge with this control type is the design of the network nodes and weights; this process is a specialty unto itself and requires extensive testing and system identification to tune the layers for appropriate outputs. If the controller design and testing can be accomplished, however, neural network control could be a very robust approach to hemodynamic management.

Finally, rule-based controllers are tailor-made rule sets, usually developed using expert guidance, that dictate what outputs to generate based on the inputs received to the controller. The complexity and capabilities of these controllers are limited only by the rule set, and thus they are more than capable of managing MIMO applications. Rule-based controllers, however, are particularly in need of extensive testing and validation because of the arbitrary nature of their design to assure stability and robustness of control.

Regulatory Considerations

There are no commercially available devices for management of fluid administration in clinical care at this time. All of the systems described or studied to date have been experimental. Figure 17.4 shows an experimental closed-loop fluid administration system in use during a clinical study. The subcomponents of the closed-loop system (sensor, controller, and intervention) in this case are encompassed in separate devices, as is often the case in prototype systems. It is unlikely that a setup like this would be approved for commercial use, however, given the risks involved with data transfer between the separate devices.

Fig. 17.4
figure 4

Example experimental closed-loop fluid administration system. The system components are configured linearly from right to left in the figure above, and each component in this system is made by an individual manufacturer. Data are captured by the monitor, passed to the controller through a USB cable, analyzed, and an action determined and then passed through another USB connection to the fluid pumps, which are activated or disabled as appropriate

Full size image

Regulatory requirements will in fact be one of the principle challenges faced by closed-loop devices in the years to come. In the case of fluid administration, one benefit is that the safety margin is quite large when compared to, for example, propofol delivery via closed loop. The flip side is that the clinical effect of propofol is relatively short-lived, whereas once IV fluid is administered, it is difficult to “remove” except via diuresis.

Specific guidelines have been issued in relation to closed-loop medical devices by the International Electrochemical Commission [23]. In addition to meeting these requirements, closed-loop devices will be expected to meet high safety standards before approval can be expected. Despite these challenges, however, the long-term gains in safety and standardization will benefit patients and providers alike.

Control of Blood Pressure

In addition to fluid management and cardiac output optimization, there is a definite role for closed-loop control of inotropic and vasopressive agents in hemodynamic management. A comprehensive system would strive foremost for optimization of blood volume, using other interventions to temporize while blood volume is restored or to optimize blood pressure or oxygen delivery once volume is adequate.

A variety of studies over the years have performed closed-loop control of hemodynamics through the use of pharmacologic agents. Over 30 years ago closed-loop administration of nitroprusside for the management of hypertensive crises was discussed [24]. Other studies have since looked at nitroprusside [25–28], management of circulatory shock [29–32], vasopressor administration [33–35], and management of blood pressure during and after cardiac surgery [36, 37].

Conclusion

Closed-loop fluid management and hemodynamic optimization is not a new concept, but it is a concept whose time may have finally come. There is a great deal of work to be done in identifying appropriate combinations of parameters to monitor, learning how best to deliver fluid and pharmacologic interventions, and even in determining optimal control strategies for clinical care areas. Now that closed-loop control is a possibility in this arena, however, it is also an almost certain component of future clinical management.