Closed Loop Control And Anesthesia Biology Essay

Published: Last Edited:

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

It is the mission of an anesthesia provider to ensure the patient's safety while providing satisfactory surgical conditions. To achieve these goals, the anesthesia provider selects a combination of available agents and techniques. Determinants of initial dose are based on the information collected during the preoperative interview. Anesthesia providers observe the patient's response to initial interventions and adjust the anesthetic plan accordingly. Subsequent interventions will be further tailored, based on observation of the patient's status and previous responses to keep the patient in a desired state or as close to that state as possible. In control engineering terminology, this establishes a closed-loop control system because of the connection between the responses and the interventions of the anesthesia provider. Compared with other common closed-loop control systems, one thing is very special here: the controller is human.

As technology advances, more and more physiological parameters can be monitored. The use of electrocardiography, blood pressure, temperature, pulse oximetry, capnography and neuromuscular blockade monitoring has reduced patient morbidity and mortality and revolutionized anesthesia practice. Arterial lines, central venous pressure, cardiac output/cardiac index, and bispectral index have gained popularity in the operating room. All these monitors provide an early warning of acute physiologic deterioration before irrevocable damage; however, it also enhances the need for anesthesia provider vigilance. Humans, by nature, often struggle to pay continuous attention to all those monitors simultaneously.

We live in a rapidly changing world. Technology is altering every aspect of our lives, and it is quite evident that today's operating room is a very complex environment. The anesthesia provider's ability to multitask continuously under difficult circumstances is one of the most important features of modern anesthesia. Anesthesia providers are often overloaded with information and multitasking necessities in an extremely complex work environment.

Noise, temperature, humidity, exposure to anesthetic agents, and ambient lighting can affect the anesthesia provider's performance (Wong, Smith, & Crowe, 2010). Multiple players from different professional backgrounds work in the operating room at the same time. Each individual has a set of goals, abilities, and limitations. Interpersonal interaction among anesthesia personnel and other operating room team members often dominates the work environment and becomes another source of distraction.

Human performance is the most important component of every anesthetic regimen and is tightly related to patient safety. To err is human. It is unreasonable to expect error-free human performance. The Institute of Medicine (IOM) showed that preventable medical errors kill more people than automobiles, breast cancer, and AIDS (Kohn, Corrigan, & Donaldson, 2000). A follow-up paper in 2005 commented that the original work of the IOM probably underestimated the seriousness of the problem (Leape & Berwick, 2005). Consequently, this huge problem is concerning to all parties involved in health care, and must be addressed.

In contrast to monitoring by a human being, an automated closed-loop controller has some obvious benefits. First, it can provide continuous attention to all the monitoring devices. Second, it will not be distracted from its tasks. Third, its algorithm is 100% repeatable, meaning it can be tested and reviewed easily to improve its performance. Fourth, in most situations, it can control devices more accurately than human hands. Most important of all, it will automate a lot of manual tasks and allow anesthesia providers to pay more attention to the patient and the surgical field.

The population of industrialized nations is aging so that many more surgical candidates are older than before. At the same time, the progress in technology and medicine has allowed for increasingly complex operations to be performed on increasingly ill patients. The physiologic reserve decreases with age and significant comorbidities. These surgical candidates need more and continuous attention because their tolerance to stress is low. With a growing demand for more complex surgical procedures in patients with limited physiological reserves, the need to fine-tune anesthetic management is greater than ever. Accurate and frequent adjustment made by the anesthesia provider can produce clinical benefits but requires clinical expertise and is a labor-intensive process that may divert attention from critical actions, resulting in paradoxically suboptimal outcome or even threatening patient safety. A well-designed automated controller or computer-based controller can assist anesthesia providers to perform the many required tasks in a less demanding manner and be very helpful to anesthesia providers.

Open-loop Control System vs. Closed-loop Control System

Anything that manages, directs, or regulates the behavior of other devices can be called a control system. Control systems can be broadly classified as either closed-loop or open-loop, the major difference being if the system uses feedback from the controlled process to adjust its input to the system.

In an open-loop system, the output of the controlled process is not measured or does not provide any information back to the controller for adjusting the input signal. The process is controlled directly, and only, by a fixed input signal. One common example of an open-loop control system is a washing machine. The motor is only under the control of preprogrammed input signal. The output of the washing machine, the cleanness of the cloth, is not measured by the system. Neither does it have any effect in controlling the machine.

Another example of an open-loop control system is a target-controlled infusion (TCI) device. The controller, often a microprocessor, performs the complex calculations and adjusts the rate of infusion based on the desired plasma concentration and a set of pharmacokinetic models. Neither the real plasma concentration of the medicine nor the effect of the medicine is measured or used to adjust the rate. A clinician using a TCI system to administer drugs often sets a desired drug concentration first and then adjusts it based on clinical observation of the response of the patient. To close the loop, a human is required to link the effect-site concentrations displayed by the TCI system and drug effect information shown by the effect monitors.

In a closed-loop system, the output of the controlled process is closely monitored and compared to a reference point. This information is used to modify the input signal. Although the term "closed-loop control" is foreign to some, it has a long history. In about 270 BC, the Greek Ktesibios designed an automatic device for water clocks (Lewis, 1992). A water clock is one of the oldest time-measuring instruments. It has a smaller tank that held water dripped, very slowly, from a larger reservoir, raising a float and moving the time pointer. To make sure the clock kept proper time, the larger reservoir has to be filled precisely to the right level. A slave had to be assigned to refill the reservoir constantly. Ktesibios's automatic regulator used a float in the reservoir to measure the water level and linked it to a valve that would let more water in when the water level dropped below the designated level. This device freed the slave from the water clock and made it affordable for common people. A similar device is still used in a modern flush toilet.

Another landmark in the history of closed-loop control systems is James Watt's steam engine. Invention of the steam engine marked the accepted beginning of the industrial revolution, affecting almost every aspect of daily life. Robert E. Lucas Jr. (2002), a Nobel laureate, said, "For the first time in history, the living standards of the masses of ordinary people have begun to undergo sustained growth . . . Nothing remotely like this economic behavior has happened bef ore." However, James Watt's steam engine was not the first steam engine. The main difference between James Watt's steam engine and earlier ones is the control system. The earlier engines were regulated manually and were very inefficient, implying they were not suitable for industrial use. Watt designed a centrifugal flyball governor, which regulated the speed of his engine automatically and launched the industrial revolution. In 1868, J. C. Maxwell analyzed the stability of Watt's flyball governor using differential equations, signifying a new era of control theory by making mathematics the official language of control theory.

A modern everyday example of a closed-loop control system is an air conditioner. The thermometer measures the room temperature constantly and compares it to the set point. If the difference is out of the allowed range, the controller will adjust the heater or cooler appropriately. Although it would be possible to manually adjust the heater/cooler, it would be very labor intensive and require many adjustments every hour to even modestly approximate the performance that an automated controller easily achieves.

Most living organisms, including humans, use closed-loop systems to control critical physiology processes. For example, human blood pressure is closely monitored by the body. If it is off from the set point, the body will use many mechanisms to correct it and move it back to the set point or as close as possible. Our body temperature is also under the control of closed-loop systems.

On the one side, an open-loop controller is simple to design and costs less. It is often enough for simple processes where feedback is not critical. It is also useful for well-defined systems where the relationship between the input and the resultant state can be modeled exactly. It is obvious that anesthesia does not meet any of these conditions. In complex processes or where feedback is critical, open-loop control systems are not as commonly used as closed-loop control systems because they cannot correct or compensate for any disturbances in the system.

Additionally, a closed-loop controller is self-adjustable. Even if it is more difficult to design and, in most cases, more expensive, only a closed-loop system should be used for anesthesia, either by a "human controller" or some automated closed-loop controller, because the effects must be monitored and the interventions need to be adjusted accordingly.

Frequently Used Controllers in Health Care

A closed-loop system has to have some basic components, including a process, controlled variables, controller, and actuator. The process is the target the controller designed to control. The controlled variables are often measureable indicators of the process. The controller is the "brain" of the system. The actuator carries out the command from the controller. These can be easily mapped to anesthesia. Anesthesia itself or the depth of anesthesia is the process which we want to control. The controlled variables, also called target variables, are the physiological parameters, such as vital signs and other indicators of anesthesia depth. The controller is the anesthesia provider or any other possible automatic system. The actuator is the gas delivery system, pump, or any other drug-delivery tools.

The type of controller used will have a significant impact on the reliability and limitations of the whole control system. The decision is mainly dependent on the characteristics of the target process. Many types of closed-loop controllers have been used in the medical field. The most common ones will be discussed here. These controllers can be measured objectively using parameters that measure the responsiveness of the controller to an error, the amplitude of the error, and the degree of oscillation. Another benefit of these automatic controllers is that they can be trained or tested by a simulation process. This is very important for applications in the health care field since testing controllers on patients are very expensive or, sometimes, unethical.

The proportional-integral-derivative (PID) controller is the most commonly used closed-loop controller in the industrial world. It uses three separate parameters to adjust the input to the system: the proportional, the integral, and the derivative values, denoted P, I, and D. P depends on the present error. I depends on the accumulation of past errors. D is a prediction of future errors, which is based on current rate of change. To reduce system oscillations, tuning must take place to adjust the gain of each of these three components.

In the absence of knowledge of the underlying process, a PID controller is the best choice. Most times, by tuning the three parameters in the PID controller algorithm, the controller can provide satisfactory control. However, the tuning itself may need a long time and a lot of data. In many given situations, it may not provide optimal control. PID controllers can be used as the sole control method in clinical applications but more often are used in combination with other controllers that have further advantages over PID.

The model-based controller is based on what is known about the process. A mathematical model is built first based on previous data. The model-based controller then uses the model to calculate the required input for a specific level of output. All current models are based on some assumptions, usually simplifying the reality to make a mathematical model possible. Some of these assumptions are just not true so, not surprisingly, the predictive accuracy of current models is imperfect. However, it is very obvious that this type of controller will be more efficient than the PID controller if the model is reasonably right.

Pharmacokinetic models and pharmacodynamic models are the most used models in the medical field. A thorough understanding of the dose-response relationship is essential to achieve the specific therapeutic drug effect while minimizing side-effects. The relationship between drug effect-site concentration and clinical effect has to be integrated into the model to close the loop. A way to do this was recently reported and showed excellent performance in a simulation study (Hahn, Dumont, & Ansermino, 2011). One big shortcoming of this controller is that most models are based on average data from a specific population. Caution should be applied when extrapolating and using the models in groups different from the original validating population, especially in children, the elderly, and patients with significant comorbidities. Even for patients within the original validating population, for any specific individual, there is always a deviation from the population average. The controller has to specify a mechanism to correct this deviation.

An Artificial Neural Network (ANN) is an information process system inspired by the structure of our brain, a biological network of interconnected neurons. Just like our brain can learn from the outside world, the ANN changes its structure based on external or internal information that flows through the network during the learning phase. The ANN can perform tasks that a linear program cannot. The ANN is parallel in nature, so it can continue to work well even if one node of the neural network fails. The ANN network learns and modifies itself just like our brain and does not need to be reprogrammed. One disadvantage of the ANN is it requires a long processing time or very powerful processors when the network becomes large.

The fuzzy logic rule-based system is becoming popular, even though its name "fuzzy" sounds a little skeptical. Reality is always fuzzy instead of perfect. Humans have no problem processing fuzzy information. A lot of the human's daily language is fuzzy. For example, in the operating room, we do not have any problem in understanding that the patient's blood pressure is "a little bit low." Nonetheless, computer systems are not able to handle that concept and the machine can only execute clear-cut instructions. Fuzzy logic was first proposed by Lotfi A. Zadeh of the University of California at Berkeley in 1965 to describe a system "too complex or too ill-defined to admit of precise mathematical analysis." It can help computers with logic involved with fuzzy concepts-concepts that cannot be expressed as "white" or "black" but "gray."

In a simplified way, a fuzzy logic controller can be created via three steps. First, the membership values has to be created (fuzzify). In another word, we have to divide each set of data into ranges instead of exact values. Second, the rule table must be specified. This determines the output based on the range of input. Third, the procedure for defuzzifying the results has to be identified. When information enters the system, it will be mapped on the fuzzy table. Then a result will be generated for each appropriate rule in the rule table. The final step will convert the combined result back into a specific control signal to actuators.

The fuzzy logic controller is very robust and can be easily modified. It also can handle multiple input and output sources. It is usually very quick and cheap to implement. The disadvantage is also obvious: it needs prior knowledge to fuzzify data and set up the rule table. There is a degree of arbitrariness that goes into the design of the controller. Even when experts are used in the design process, there may be different opinions about boundaries and significance.

Literature Review of Closed-loop Control Systems in Anesthesia

Anesthesia is a very broad field. It can be divided into five aspects: hypnosis, analgesia, neuromuscular blockade, mechanical ventilation and fluid management. Hypnosis, analgesia, and immobilization are the three major components of anesthesia. Fluid management and ventilation management are also the anesthesia providers' critical tasks in the operating room. Closed-loop control systems in each of these aspects will be discussed. In each aspect, the controlled variables will be discussed first. The research related to closed-loop systems in that aspect then be reviewed.

Closed-loop Control Systems Related to Hypnosis

Hypnosis is one critical component of anesthesia. One final goal of anesthesia is to prevent awareness without overloading the patients with potent drugs. Before discussing closed-loop control systems for hypnosis, the measurements for hypnosis have to be discussed. These parameters are particularly important for total intravenous anesthesia (TIVA), where anesthesia providers do not have the advantage of a MAC value to help titrate the drugs.

One group of measurements for hypnosis is the electroencephalogram (EEG) and derived indices. Since consciousness is mainly a function of the brain, the EEG is nature's measure of hypnosis. However, the raw EEG waveforms are just complex small voltage deflections, which are extremely difficult to interpret. Modern EEG monitors obtain and process raw EEG signals over a period of time and display the information in different forms, such as compressed spectral array (CSA), spectral edge frequency (SEF), median frequency (MF), bispectral index (BIS), entropy, narcotrend index, patient state index (PSI) and cerebral state index (CSI). All these modalities have been shown, not surprisingly, to correlate with the depth of anesthesia.

The CSA is still hard to comprehend or quantify, so it is not commonly used. The SEF is usually expressed as SEF x, which stands for the frequency below which x percent of the total EEG power is located. Commonly used SEF is SEF 95. MF is just SEF 50.

The BIS is a combination of information from time domain, frequency domain, and high order spectral subparameters. The underlying algorithm is proprietary. The unique thing about the BIS is that its algorithm is based on a large volume of clinical data. It has been used and validated in many studies. A systematic review showed that the BIS is useful in improving anesthetic delivery and postoperative recovery (Punjasawadwong, Boonjeungmonkol, & Phongchiewboon, 2007). In 1996, the Food and Drug Administration (FDA) recommended the use of the BIS to monitor the depth of anesthesia to reduce the incidence of intraoperative awareness.

However, the BIS has its own shortcoming. Like all other calculated indices, the BIS has a time delay before the state can be updated, which varies from 24 seconds to 122 seconds (Zanner, Pilge, Kochs, Kreuzer, & Schneider, 2009). Erroneous placement or reduced adherence of electrodes can cause falsely elevated BIS. Since the frequency limits range of EMG and EEG signals are very close, some EMG signals can be erroneously interpreted as EEG signals, leading to a false elevation of BIS. Many electrical devices can cause artifacts in the BIS. The BIS is also not reliable if N2O or ketamine is used for anesthesia (Hans, Dewandre, Brichant, & Bonhomme, 2005; Park, Hur, Han, Kil, & Han, 2006; Rampil, Kim, Lenhardt, Negishi, & Sessler, 1998). Opioids usually function at subcortical structures that are not detected by the EEG. Clinically, the hypnotic effect of propofol is enhanced by opioids, the BIS does not show this increased hypnotic effect (Lysakowski, Dumont, Pellegrini, Clergue, & Tassonyi, 2001). Entropy, narcotrend index, PSI and CSI are not widely used and do not show any advantage compared to the BIS.

A novel member in this family is WAVCNS, which is measured by the NeuroSENSE monitor (NeuroWave Systems Inc., OH). The WAVCNS correlates with the BIS almost perfectly during periods of steady state despite fundamental algorithmic differences. During induction and emergence, the WAVCNS offers faster tracking of transitory changes with an average lead of 15-30 seconds (Zikov, Bibian, Dumont, Huzmezan, & Ries, 2006). Among the BIS, entropy and WAVCNS, the WAVCNS is the only one that can be fully modeled as a linear time invariant transfer function, which means it is a better target variable for a closed-loop system (Bibian, Dumont, & Zikov, 2011).

Another group of hypnosis measurement is based on evoked potentials (EPs). The EPs show the responses of localized areas of brain to specific stimuli. It is obvious that all EPs depend on intact neural pathways. Commonly used ones are the somatosensory evoked potential (SSEP), visual evoked potentials (VEP), auditory evoked potential (AEP), middle latency auditory evoked potentials (MLAEP), and auditory evoked potential index (AEPI). However, its complex setup and not-easy-to-interpret characteristics limited the usage of these measurements.

No monitoring system has been proven to measure the depth of hypnosis reliably for all patients and all anesthetic agents. All previously mentioned monitors can only measure the hypnotic state at the time of measurement, and none can predict if the depth of hypnosis is sufficient for the next surgical stimulus. Many researchers have been managed to build closed-loop systems for hypnosis based on these non-perfect monitoring systems.

Mayo and Bickford et al. (1950) developed a closed-loop anesthetic delivery system that regulated the administration of ether or thiopental using EEG activity. In 1984, another closed-loop system was developed to control the delivery of oxygen, anesthetic agent, and N2O (Hayes, Westenskow, East, & Jordan, 1984). This controller is based on a PID algorithm, and the target variables are inspired oxygen concentration and end-tidal enflurane concentration. This system worked well in a group of seven dogs. However, no clinical studies data are available.

Morely et al. (2000) evaluated the performance of a closed-loop control system in the administration of general anesthesia in 100 patients, either through the infusion of a propofol/alfentanil mixture or through the use of an isoflurane/nitrous oxide-based technique. The closed-loop control system worked well in clinical practice but did not demonstrate a clinical advantage over the undoubtedly attentive anesthetists in the manual control groups. The tested system is based on a PID controller. Each controller is unique even if they are based on the same algorithm. These results are hard to generalize to other controllers.

A recently patented controller for volatile anesthetics has a cascade structure with outer and inner control loops. The outer loop is a standard PID controller, and the inner controller has a model- based state feedback design. The outer control loop adjusts the end-tidal volatile concentration to obtain a desired BIS concentration. The inner control loop adjusts the vaporizer to obtain the desired end-tidal concentration. This controller outperformed a group of anesthesia providers in a small trial (23 patients) (Locher et al., 2004).

Volatile anesthetics have been used less frequently under conditions of automated control. Propofol is more popular in this field. In 1998, a closed-loop propofol delivery system was successfully used to sedate ten patients undergoing elective orthopedic surgery with spinal anesthesia (Mortier, Struys, De Smet, Versichelen, & Rolly, 1998). In this system, the propofol administration was controlled by a patient individualized adaptive model-based controller that incorporated target-controlled infusion technology combined with a pharmacokinetic-dynamic model.

A closed-loop anesthesia delivery system (CLADS) is a patented propofol delivery system that uses the BIS as the controlled variable and a standard infusion pump as an actuator. A randomized controlled trial in forty generally healthy patients showed that the CLADS is both effective and efficient compared to manual control (Puri, Kumar, & Aveek, 2007). CLADS used smaller amounts of propofol and maintained BIS to within +/-10 of target for a significantly longer time. Postoperative recovery in the CLADS group was faster. Another RCT with forty-four adult ASA 2-3 patients undergoing elective open-heart surgery reached a similar conclusion (Agarwal, Puri, & Mathew, 2009). Hemodynamic stability was better in the CLADS group, and the cumulative dose of phenylephrine used was significantly higher in the manual group. It also outperformed the manual group in postoperative sedation after open-heart surgery (Solanki, Puri, & Mathew, 2010). The CLADS also performed well in some clinically extreme situations, such as anesthetic management of resection of pheochromocytoma (Hegde, Puri, Kumar, & Behera, 2009).

Improved anesthetic agent delivery system (IAADS), a modification of CLADS, is designed to deliver inhalational anesthetics and propofol through closed-loop control with the BIS as the controlled variable. It is the first system that has been developed to control both intravenous and inhalational anesthetic agents in a closed-loop model using the BIS. This system was tested in forty patients undergoing elective cardiac surgery with cardiopulmonary bypass (CPB) (Madhavan, Puri, & Mathew, 2011). Propofol infusion was used for induction and during CPB. Isoflurane was started after intubation and stopped during CPB and was restarted after separation from CPB. Compared with the manual group, this closed-loop system used much less propofol and isoflurane and was able to maintain the BIS within target for a significantly longer period during the surgery.

A neural network adaptive system was published and successfully used to control propofol delivery for noncardiac surgery patients to maintain a desired constant depth of anesthesia (Haddad, Bailey, Hayakawa, & Hovakimyan, 2007). This system is one of the few systems that can be used during the induction of anesthesia.

In a randomized controlled trial, Liu et al. (2011) developed a PID controller guided by a BIS monitor, which allowed the closed-loop coadministration of propofol and remifentanil during both induction and maintenance of general anesthesia. Eighty-three patients were assigned to the closed-loop control group and eighty-four patients were assigned to the manual control group. The closed-loop group made more frequent but smaller adjustments to the propofol and remifentanil infusion rates. Overshoot (BIS <40), undershoot (BIS >60), and burst suppression ratios were all significantly less common, and the time to extubation was shorter in the closed-loop group. This system is among the first to control both hypnotic and analgesic administration. The study took place in multiple hospitals and involved patients that required a range of surgical interventions, many of them having significant comorbidities.

This system is not a true dual-loop system because it uses BIS as the control variable for both propofol and remifentanil. The interaction between remifentanil and propofol is not modeled in the controller. As discussed in the PID controller section, Liu's controller is eventually able to find combinations of propofol and remifentanil infusion rates that limit the error in BIS, but the path may be suboptimal. A model-based adaptive controller or other advanced controller may offer better results in the future.

Closed-loop Control Systems Related to Analgesia

Analgesia is another important component of anesthesia. Pain is one of the most unpleasant sensations, and it is more complicated than other somatosensory modalities such as touch and vibration. Pain, by nature, is subjective and is easily changed by a change in mental state. In conscious patients, pain can be easily assessed by direct communication; however, this is not the case for patients who are under general anesthesia.

In contrast to hypnosis, there is no good surrogate parameter for analgesia in anesthetized patients. Parameters of autonomic response, such as heart rate and blood pressure, have been used for a long time as signs of nociception. Sudden hypertension/tachycardia and sweating may indicate inadequate analgesia. However, these parameters are not specific. A wide range of events in the operating room such as dehydration, hypoxia, hyperthermia, massive blood loss, etc., can lead to similar effects.

Various measures of the status of the autonomic nervous system have been studied, such as skin conductance (SC), heart rate variability (HRV), and photoplethysmography variability (PPGV). These parameters, to a less extent, are still affected by autonomic neuropathy. SC fluctuates as the status of the palmar and plantar sweat glands changes, which is under the control of the sympathetic nervous system. It may be a useful method to monitor perioperative stress (Storm et al., 2002).

HRV is a physiological phenomenon where the time interval between heartbeats varies. The SA node of heart receives several different inputs, and HRV is the result of these inputs. It is measured by the variation in the beat-to-beat interval. PPGV is considered to be a good surrogate to HRV (Lu, Yang, Taylor, & Stein, 2009). The component of HRV at respiratory frequency is named respiratory sinus arrhythmia (RSA). Several studies had suggested that RSA could be used as an indicator of anesthesia (Loula, Jantti, & Yli-Hankala, 1997; Pomfrett, Barrie, & Healy, 1993; Pomfrett, Sneyd, Barrie, & Healy, 1994). The success of these parameters has been variable (Luginbuhl, Rufenacht, et al., 2006; Luginbuhl, Ypparila-Wolters, Rufenacht, Petersen-Felix, & Korhonen, 2007; Seitsonen et al., 2005).

A multivariate surgical stress index (SSI) has been proposed to measure analgesic effect. The SSI is based on a sum of the normalized pulse beat interval (PBI) and the photoplethysmographic pulse wave amplitude (PPWA). Compared with standard clinical practice, using the SSI to titrate remifentanil during surgery resulted in reduced remifentanil usage, improved hemodynamic stability (Chen et al., 2010).

Since the selection of the controlled variable remains difficult, it is hard to design an automatic system for automated administration of an opioid. Only few systems were discussed in literatures. Schwilden et al. (1993) developed a system for the automatic administration of alfentanil during alfentanil-nitrous oxide anesthesia using the EEG median frequency as a controlled variable. The detail of the controller is not clear. Luginbuhl et al. (2006) developed a model-based controller for closed-loop administration of alfentanil guided by mean arterial blood pressure and predicted plasma alfentanil concentrations. Both systems provided appropriate doses of alfentanil to patients in small clinical trials. Neither of them was compared to human group. The interactions between agents were also not well analyzed.

Closed-loop Control Systems Related to Neuromuscular Blockade

There is no doubt that neuromuscular blockade (NMBD) is an important part of general anesthesia. Mencke et al. (2003) showed with a double-blind controlled study that the use of neuromuscular agent improved the quality of intubating conditions and decreased postoperative hoarseness and vocal cord damage. Naguib M. et al. (2001) suggested that even the placement of laryngeal mask airway can be facilitated with low-dose rocuronium.

It has been shown that NMBDs can improve surgical conditions, especially in abdominal surgery. For patients who cannot tolerate deep anesthesia, NMBD can be used to maintain a lighter plane of anesthesia and provide satisfactory surgical condition at the same time. For procedures where even slight movement could result in critical events, profound NMBD is often necessary.

The ideal goal of titrating neuromuscular agents is to provide (if necessary) sufficiently profound neuromuscular blockade during surgery and immediate reestablishment of normal neuromuscular transmission at the end of surgery. Sometimes this is tricky to achieve in the OR. At the same time, the effects of neuromuscular blockade agents are relatively easy to measure. Both of these make neuromuscular blockade a natural target for closed-loop controllers. A subjective (qualitative) visual or tactile assessment of a response to peripheral nerve stimulation is the most common method of neuromuscular monitoring used in the OR. It cannot be used in an automatic closed-loop system. Many quantitative neuromuscular monitoring techniques have been established: electromyography (EMG), mechanomyography (MMG), acceleromyography (AMG), kinemyography (KMG), and phonomyography (PMG).

EMG, the oldest method of neuromuscular monitoring, is based on recording of the compound action potential after evoked stimulation. MMG measures isometric contraction of a peripheral muscle (usually the adductor pollicis) in response to ulnar nerve stimulation. AMG measures acceleration of a given target when moved by a muscle. KMG uses a piezoelectric polymer sensor to detect the degree of bending produced by the thumb in response to electrical stimulation of the ulnar nerve. PMG depends on the sounds that a muscle contraction evokes, and the sound intensity is proportional to the force of contraction. A distinct microphone is placed alongside the monitored muscle to record the sounds from the isometric muscle contractions.

EMG and MMG are the traditional ways to monitor neuromuscular blockade. They are used in many researches but much less frequently clinically because of the relatively elaborate setup and bulky equipment. AMG and KMG are commercially available, easy to apply, and relatively inexpensive. PMG is very promising but still not commercially available. At present, AMG is "the best compromise with respect to ease of use, practicality, versatility, precision, and applicability at various muscles" (Hemmerling & Le, 2007). However, one study showed that, compared with AMG, EMG is better for a closed-loop control system and more reliable for use in daily practice as it is less influenced by external disturbances (Hanzi, Leibundgut, Wessendorf, Lauber, & Zbinden, 2007).

Various closed-loop systems for muscle relaxants have been proposed and performed well. Cass et al. (1976) used a computer to control the injection of d-tubocurarine, gallamine, alcuronium, or pancuronium and successfully reduced the electromyogram to a preset level for one hour in sheep.

The early closed-loop systems for vecuronium (de Vries, Ros, & Booij, 1986) and atracurium (Wait, Goat, & Blogg, 1987) were just simple on-off systems. Then closed-loop systems based on PID controllers were developed for atracurium (O'Hara, Derbyshire, Overdyk, Bogen, & Marshall, 1991; Webster & Cohen, 1987). Later, a model-based controller with an internal pharmacokinetic-pharmacodynamic model of the patient was reported (Schwilden & Olkkola, 1991). More recently, complex algorithms, such as the self-learning fuzzy logic controller and hierarchical fuzzy logic controller, were investigated too (Ross, Mason, Linkens, & Edwards, 1997; Shieh, Fan, Chang, & Liu, 2000). Some controllers can remain useful even in the presence of disturbances that can arise in routine clinical conditions, such as additional manual bolus, turned off/on, and empty infusion bag (Eleveld, Proost, & Wierda, 2005).

All these controllers, even the early simple ones, can provide stable control of neuromuscular blockade despite the considerable individual variation in neuromuscular block requirements among patients. Most of these controllers were only tested with one neuromuscular blockage agent. One study showed that the differences among the performances of a model-based controller for the administration of atracurium, mivacurium, rocuronium, and vecuronium in 159 adult surgical patients are clinically insignificant (Kansanaho & Olkkola, 1996).

Despite the abundance of these controllers and the excellent performance, they are still not popular in the clinical setting. One reason is that many controllers are not easy to set up, and this makes them unsuitable for routine clinical use. Another reason is the recent appearance of drugs such as cyclodextrin that can reverse rocuronium-induced neuromuscular block rapidly and completely (de Boer, van Egmond, van de Pol, Bom, & Booij, 2006). When overdose carries little risk and the effect can be rapidly reversed, the clinical interest of automatic closed-loop control in this field is reduced.

Closed-loop Control Systems related to Mechanic Ventilation and Oxygenation

Since mechanical ventilation is such a specialized field, only a very short discussion is provided here. For ventilation and oxygenation, the target variables are easily identified: end-tidal CO2 (EtCO2) and oxygen saturation (SpO2). The monitoring devices for both are validated and widely used in everyday practice. These devices can provide continuous monitoring.

The seminal study published more than half a century ago pointed out the possibility to automatically adjust the ventilatory support according to the patient's changes in respiratory mechanics and ventilatory demand (Saxton & Myers, 1957). In this paper, the authors described a servomechanism to automatically adjust the EtCO2 by regulating the negative pressure of an iron lung ventilator.

Traditional ventilation modes such as volume controlled ventilation (VCV), pressure controlled ventilation (PCV), synchronized intermittent mandatory ventilation (SIMV), and pressure support ventilation (PSV) are used in the operating room every day all over the world. Many advanced modes are commercially available today: proportional assist ventilation (PAV), neurally adjusted ventilatory assistance (NAVA), and adaptive support ventilation (ASV). PAV and NAVA are basically advanced versions of PSV. ASV is a mixture of PSV, PCV, and SIMV. All these advanced modes have been studied in ICU settings and showed some advantages (Arnal et al., 2012; Colombo et al., 2008; Xirouchaki et al., 2008). No available study aimed to compare these different modes. Their usage in the OR has not been investigated. More data is still needed to show clear clinically significant advantages. All these modes focused on the support of respiratory muscle and control EtCO2.

An ideal fully closed-loop system should finish two tasks automatically: adjust tidal volume and respiratory rate based on the EtCO2 and adjust the FiO2 and positive end-expiratory pressure (PEEP) to keep the SpO2 within target range. A commercially available fully closed-loop control system of ventilation and oxygenation (IntelliVent-ASV) was investigated recently (Arnal et al., 2012). This randomized crossover study showed that this system was safe and produced the same results in terms of oxygen saturation but with less pressure, volume, and FiO2. Further study is needed to show if this system can improve clinical outcome in patients.

Mechanical ventilation is in an especially advanced era. Some researchers even suggested that closed-loop ventilator management may be an emerging standard of care (Wysocki & Brunner, 2007).

Closed-loop Control Systems Related to Fluid Management

Fluid management is an integral and important part of the perioperative care. Many postoperative complications are related to giving the wrong amount of intravenous fluid in the operating room.

The traditional fluid management is based on standardized formulas, which purport to account for preoperative fluid losses, ongoing maintenance requirements, intraoperative blood loss, and insensible losses. The rates are frequently titrated to obtain a urine output of 0.5 to 1 mL/kg/h. There is substantial variation of both opinions and practice regarding perioperative fluid resuscitation (Chong et al., 2009).

The modern approach to fluid management is based on the concept of goal-directed therapy. The final goal of fluid management in the operating room is to optimize stroke volume and cardiac output. A meta-analysis of about 5,000 patients showed that patients in the goal-directed fluid management group had a lower risk of pneumonia, renal complications, and a shorter length of hospital stay (Corcoran, Rhodes, Clarke, Myles, & Ho, 2012).

The most difficult part of closed-loop fluid management is to find the control variables. Hypotension, tachycardia, and oliguria are common signs of hypovolemia. They are late signs and not specific. This means that mean blood pressure, heart rate, and urine output are not ideal control variables for closed-loop fluid management system. The central venous pressure (CVP) is widely used to guide fluid therapy in hospitalized patients. A recent systematic review showed that the CVP is not a reliable indicator of fluid responsiveness and should not be used to make clinical decisions regarding fluid management (Marik, Baram, & Vahid, 2008).

The pulmonary artery occlusion pressure (PAOP) and cardiac output measured through the pulmonary artery (PA) catheter may be the best way to predict fluid responsiveness. The invasive nature limited its usage. At the same time, the benefit of PA catheter is still controversial (Harvey et al., 2005; Harvey et al., 2006).

The esophageal Doppler is a minimal invasive way to monitor cardiac output. Several studies showed that goal-guided fluid management based on esophageal Doppler cardiac output monitor is related to better patient outcome (Kuper et al., 2011; Wakeling et al., 2005).

Systolic blood pressure variation (SPV) is a sensitive indicator of hypovolemia, and it even reacts earlier than the CVP (Coriat et al., 1994; Perel, Pizov, & Cotev, 1987). The arterial pulse pressure variation (PPV) is a better fluid responsiveness indicator than the SPV in septic patients with acute circulatory failure (Michard et al., 2000). In patients after anesthesia induction, PPV works much better than CVP in assessing volume status (He et al., 2011). Both SPV and PPV can be easily calculated from arterial pressure waveform.

Noninvasive assessment of fluid responsiveness is also possible. Respiratory variations in pulse oximetry plethysmographic waveform amplitude (DeltaPOP) can predict fluid responsiveness in the operating room (Cannesson et al., 2007). The pleth variability index (PVI), an automatic and continuous monitor of DeltaPOP, can predict fluid responsiveness noninvasively in mechanically ventilated patients during general anesthesia (Cannesson et al., 2008).

SPV, PPV, DeltaPOP, and PVI require mechanical ventilation in closed chest, tidal volume of at least 6 mL/kg, and no arrhythmia. In a spontaneously breathing patient, changes in intrathoracic pressure might be insufficient to modify the loading conditions of the ventricles so these indicators do not work well (De Backer & Pinsky, 2007). Some recent studies suggest that a deep inspiration maneuver or valsalva maneuver may help the performance of these indicators in spontaneously breathing patients (Monge Garcia, Gil Cano, & Diaz Monrove, 2009; Preau et al., 2012). The noninvasive ones, DeltaPOP and PVI, are very sensitive to vasomotor tone.

Closed-loop fluid management is still in its early age. An automated closed-loop resuscitation system, based on a PID algorithm, can adjust infusion rate based on urinary output (Hoskins et al., 2006). This system outperformed manual adjusting in making stable urine output.

In 2011, a novel closed-loop fluid management system was reported (Rinehart et al., 2011). It was based on the PPV, but the underline algorithm of the controller was not published. In the management of simulated massive hemorrhage, compared to a group of anesthesiologists, the closed-loop system showed a higher and steadier cardiac output CO in the competition. No information about this system in real patients is available. In the future, a closed-loop system for fluid management using both cardiac output and the dynamic predictors of fluid responsiveness may provide us startling results.

Future Research

Anesthesia providers are often compared to airline pilots: induction, maintenance, and emergence are equated to the takeoff, cruising, and landing. However, closed-loop systems have been used as copilot for many years in airplanes to provide a safe, stable trip to the travelers. If we can build automated closed-loop systems to reliably keep a 350,000-pound aircraft in the air for hours, what stops us from building automated closed-loop systems for anesthesia?

Despite the potential utility, there are numerous challenges for closed-loop in anesthesia, which include, but are not limited to, the complexity of anesthesia in humans, uncertainty in ideal target variables, regulatory approval, safety, and acceptance by practitioners.

The behavior of an airplane can be accurately described with physics' laws and equations. In contrast to this, our understanding of consciousness and the mechanisms of anesthetic-induced hypnosis is far from complete. The complexity of anesthesia does not lend itself for simple controlled variables. The accuracy and validity of the variable(s) monitored by the system are extremely important to a closed-loop system because they provide feedback to the controller. Without good feedback, the controller cannot respond appropriately to changes.

Besides the controlled variables, the devices used to measure the controlled variables can cause a problem too. The selection of a monitoring device is a key strategic decision in most control systems. If possible, sensors should be linear and time invariant to ensure that that they do not add uncertainty to the system. However, with few exceptions, none of the current monitoring devices in anesthesia reached this goal. One direction for future research in this field is to search for better surrogate parameters for anesthesia and better monitoring devices.

Among many things hindering the clinical application of closed-loop systems, patient safety is a major concern. In theory, a closed-loop control system automates typical decision making of routine care and reduces the workload of anesthesia care provider. Drug administration is an asymmetrical process: we can actively give more but cannot actively take back. The automated systems itself must be fail-safe. If they malfunction or were given bad data because of artifacts or noise, they must be capable of filtering out bad data and behaving in a way that does no harm to the patient. None of the current tested closed-loop anesthesia systems has shown this via big persuasive clinical trials.

Food and Drug Administration approval can also be difficult. The base of FDA approval, for any medical product, is the risk-benefit analysis. If no meaningful clinical benefit can be demonstrated, any potential risks associated with a closed-loop anesthesia delivery system would be deemed unacceptable. One of the directions for future research is to design clinical trials which can demonstrate a meaningful clinical benefit associated with the use of the product. Trials that can demonstrate a clinically significant safety benefit, such as a reduction in adverse events related to too light anesthesia or too deep anesthesia as compared to current standard clinical practice, will increase the likelihood of FDA approval significantly. Given the very low incidence of such adverse events, an extremely large sample might be required. Trials designed for less anesthetic drug requirement and improved speed of recovery of patients are easier to carry out but less helpful too.

The clinician acceptance may also be difficult. Many anesthesia providers may believe that these devices will "take over." The reality is that these systems are not meant to run in the absence of a supervising clinician, just like you never see a commercial airplane flying itself without a pilot. Closed-loop systems are meant to be supervised by and help experts. In addition, this can standardize patient care and add another level of patient safety. Educating anesthesia providers is another major task in this field.

To date, most closed-loop systems offer only "single-input-single-output control." The interactions among drugs administered by the anesthesia provider, especially between opioids and hypnotics, are clinically very significant and have been studied in detail using response surface methods (Bouillon et al., 2004). Another direction for future research is to combine the individual systems for hypnosis, analgesia, neuromuscular blockade, fluid management, and ventilation into comprehensive controllers capable of fully integrated anesthesia management. Those effects in human body are interconnected. The clinically used agents also have effects on more than one system and often interact with each other. A single comprehensive controller will allow for cooperation among the various individual components and avoid overreaction of any of the individual system. Multiple-input-multiple-output controllers are a logical next step. Recently a double-input-double-output closed-loop system was published (Janda et al., 2011). This system controls the depth of anesthesia and neuromuscular blockade using the BIS and the EMG as control variables simultaneously. It is used to administer propofol and mivacurium in twenty-two ASA 1-3 patients to maintain anesthesia after trachea intubation and successfully maintain the target values of BIS and EMG with a high level of precision. McSleepy is the first completely automatic anesthesia delivery system (Hemmerling, 2009). The anesthesia providers have to input the patient's information, including age, height, weight, and sex; the type of surgery being performed; and the drug of choice. Through unpublished algorithm, McSleepy will monitor and control the drug doses every minute and display all relevant anesthesia data on the user interface.

Closed-loop control systems perform better in many clinical trials. It must be remembered that the best anesthesia providers will outperform the controller, particularly because of his ability to anticipate the consequence of surgical events. Although the immunity to distractions, precision of control, and freedom for anesthesia providers from labor-intensive manipulation are reasons enough to further pursue their development, their final application to clinical care will depend on the future study in this field.