Why would one want to use a closed-loop controller for drug administration? Control and feedback play a crucial role in both the natural and engineered world. Despite the prevalence of control systems, we are often unaware of their presence. In the engineered world, control is omnipresent and although rarely visible it is a critical enabling technology.1 For example, mobile telephony would not be possible without sophisticated schemes for power control, automatic frequency control, and automatic gain control. Modern cars have dozens of sophisticated control loops, from braking control, steering control, to powertrain control. Indeed if it were not for control, cars would not be able to meet today’s stringent emission control requirements. Control is also a key technology for the development of the smart grid required to supply electricity efficiently to our homes today and in the future. Homeostasis is the physiologic closed-loop control, which is present in every system in every living organism. In nature, from the complex ecology of the planet to the single cell, feedback and control are essential to maintain life.2,3
Even today the anesthesiologist is constantly engaged in control and feedback. Actions, such as adjusting drug or fluid administration, are based on some observation of the clinical environment or monitoring devices. This control can extend to hypnosis, analgesia (antinociception), neuromuscular blockade, temperature, metabolic status, ventilation, and hemodynamic homeostasis (Fig. 1).4
Due to the complexity and performance pressure of the clinical environment and the large degree of individual patient variability, the performance of the anesthesiologist controller is suboptimal. In addition, there is increasing evidence that intraoperative performance by the anesthesiologist controller may influence longer-term outcomes. Intraoperative normothermia5 and protocol-driven fluid optimization6 have been shown to reduce postoperative mortality. Occurrences of low mean arterial blood pressure (MAP) and deep hypnotic levels may be associated with postoperative mortality.7–9 A cumulative time with Bispectral Index (BIS) (an electroencephalographic [EEG] based index of hypnotic effect) <45 has been associated with poor outcomes in the elderly,10 in patients with cancer,11 and during cardiac surgery12 but may be less important in patients undergoing noncardiac surgery.13 A significant limitation of many of these studies has been the inability to keep a constant measured hypnotic effect even when defined by the study protocol.14 This highlights an opportunity to study the impact and clinical outcomes of using computer-enhanced control and feedback.
A decrease in variation in clinical practice is a key goal in quality improvement.15 Independent decision making, diversity of practice, and suboptimal human control significantly limit our ability to provide strict standards of care.16 Although one could argue that many benefits could be achieved by a real-time clinical decision support system, for continuous adjustment of drug infusion rates, it is much more efficient to use closed-loop control technology. There will always be a limitation of human performance. Indeed, a feedback controller can act at a much higher frequency and without distraction compared with a clinician, and a controller will be more powerful and accurate when compensating for dynamic interactions in a multivariable setting. In addition, as seen later, only feedback has the ability to reduce the effect of uncertainty.
Thus, appropriately designed and implemented control and feedback systems in anesthesia will provide less variability in desired clinical effects than manual adjustments performed by the anesthesiologist,17 while freeing them for higher-level clinical tasks and decisions. Optimally controlling every aspect of clinical practice will enable us to understand the impact of variation in practice, establish optimal care, and increase the safety of our patients. This must of course be done without increased risk to the patient.
Although a number of experimental systems for closed-loop control of anesthesia have been reported, there are no clinical systems available today. However, if such systems are ever to be approved by regulatory bodies, they have to be designed using rigorous control engineering and software engineering principles, and proven to be safe. For this, close collaboration between anesthesiologists and engineers will be required. To facilitate such collaboration, in this article we first give a brief, high-level introduction to key concepts in control engineering for anesthesiologists. For a nonmathematical introduction to control theory and principles, the reader is referred to the excellent book18 by Albertos and Mareels, on which the next section draws extensively. We then provide an overview of clinical studies of closed-loop control of depth of anesthesia and of analgesia and finally propose a roadmap for commercialization of a closed-loop system.
BASIC CONTROL CONCEPTS
Signals and Systems
A signal is a record over time of a variable, e.g., heart rate, temperature, or depth of hypnosis (DOH). A continuous signal recorded at fixed intervals is called a sampled-data signal, and usually stored on a computer as a time series. A system can be defined as a closed set of interacting components that exchanges information with its surrounding. From a control engineering viewpoint, a system reacts in a dynamical fashion to an external stimulation (input) received from the environment, the observed or extracted response being called the output. In control, the input is usually the manipulated variable (e.g., the drug infusion rate) while the output is the controlled variable (e.g., the DOH). Inputs are manipulated by means of actuators (e.g., infusion pumps) while outputs are extracted or measured by means of sensors (e.g., DOH monitor). On occasion, inputs cannot be manipulated but can only be observed, either directly or by the response they generate from the system. Those inputs are then called disturbances (e.g., a surgical stimulus).
A model is (usually) a mathematical representation of the dynamic behavior of the system under study. Models for control tend to be simpler than high-fidelity models developed for understanding the inner workings of a system.19 Models can be derived from first principles, typically consisting of a set of differential equations, often linearized around some operating point of the system. This set of differential equations can then either be transformed into a transfer function, that represents the input–output relationships, or into a state-space model that consists of a first-order matrix differential equation. The state of the system summarizes its past, and is such that its knowledge, combined with the knowledge of the inputs, suffices to predict the future of the system. For example, a pharmacokinetic and pharmacodynamic (PKPD) model can be represented by a state-space model where the state of the system is a vector containing the concentrations of the various compartments, or by a scalar transfer function representing the relationship between the infusion rate and the effect-site concentration, without explicitly representing the other concentrations. Commonly, for control purposes, transfer functions models are derived from data, a process called system identification.20
What is Control?
A control system is a device designed to regulate the operation of an apparatus or system. Control theory is concerned with the analysis and synthesis of control systems. Although control theory is intimately linked with feedback theory, a control system may also involve use of feedforward, automata, and real-time optimization.
Many systems can be controlled using simple on/off, open/close control. Such systems are also called sequential control systems or event-driven systems. Common examples are washing machines, automated doors, and car wash systems. On/off controllers can be used to control a continuous system when the performance requirements are not necessarily demanding. The typical example is the thermostat used to regulate room temperature. The disadvantage of such systems is that the room temperature is always oscillating around the desired temperature.21
In anesthesia, the archetypical application of open-loop control (Fig. 2A) is target-controlled infusion (TCI).22 In this application, a population model of drug distribution and effect is used to adjust the drug infusion rate, to target a hypothetical plasma or effect-site concentration that is not measured. The performance of such an open-loop system is highly dependent on the accuracy of the model on which it is based. However, because this model is rarely accurate for the patient being anesthetized, a sine qua non condition for TCI to function is for the attending anesthesiologist to adjust the target concentration until the desired clinical effects are obtained. In effect, this requires the clinician to manually close the loop. A widely used variation of open-loop control is feedforward control where a measured disturbance is used to control the manipulated variable (Fig. 2B). In an anesthesia setting, this would correspond to giving a bolus of remifentanil before a noxious stimulus to minimize a patient’s reaction before it occurs. Feedforward control also relies on an accurate model of the system and will not be appropriate in the presence of significant uncertainty, or unmeasurable disturbances (such as clinical interventions). It is thus rarely used on its own, but rather in combination with feedback control.23 That feedback can be performed by the clinician has been demonstrated by Mandel and Sarraf24 who proposed to infer the effect-site concentration at loss of responsiveness and use that value for target during maintenance of anesthesia for short, minimally invasive procedures such as endoscopy. Using computer simulations, they demonstrated (not surprisingly) that even such feedback reduces the effect of interpatient variability.
Although used since antiquity in devices such as the Egyptian water clock, and in the 19th century in the steam engine governor, closed-loop control has only been formulated as a rigorous theory for less than a century. The modern concept of feedback was rigorously formulated in 1934 by Harold Black25 with the invention of the feedback amplifier. In those early days, repeater amplifiers for telephony were built from glass tubes that had nonlinear and uncertain characteristics with significant distortion. Black25 proposed a feedback amplifier with spectacular results, showing that feedback allows the design of good systems from components with poor performance characteristics. Black’s invention would lead to a revolution in telecommunication. By analogy, one could say that feedback should allow the administration of anesthesia with consistent and predictable results despite interpatient variability.
The principle of a typical feedback loop (Fig. 2C) is deceivingly simple: the measurement of the system output (controlled variable) is compared with the desired target value (set point), the difference between the output and the set point being used to compute the value of the system input (manipulated variable) (Fig. 2C). As has been shown by Black25 and many others since, feedback has some amazingly powerful properties. A fundamental property of feedback is that it reduces the sensitivity to external disturbances and to component uncertainties. Reliable performance is not dependent on models or perfect characterization of components. This is the most significant difference between open-loop and closed-loop control. Feedback thus outperforms open-loop control when there is a significant uncertainty about the system to be controlled and when unknown disturbances are present.25 This alone is a strong argument against the use of TCI, without some form of feedback.
Feedback control is not without risk. Poor design or overaggressive tuning can lead to instability, resulting in oscillatory or runaway behavior, obviously a disastrous consequence.26 A major performance-limiting factor in feedback control is the presence of delay in the response of the system.25 Uncertainty or variability in the system’s characteristics, and especially in variability in the delay (time lag from initiation of an event until detection by a sensor), further exacerbate the reduction in performance.25
A pivotal concept in safe controller design is robustness. Robustness strives to guarantee stability and minimal performance despite the expected uncertainty in the feedback variable (such as the measured EEG effect), the model, and in the specific patient. The design of a feedback controller is then always an exercise in finding the appropriate compromise between performance and robustness. For the design of a robust controller, control engineers have at their disposal a number of formal techniques known as robust control theory.27 The use of robust control theory requires the development of a nominal model together with characterization of the uncertainty in system components (such as PKPD interpatient variability or surgical stimulation). No feedback controller should be used unless its design can be verified and both robust stability and performance can be guaranteed against a specified uncertainty. This is especially true for safety-critical applications such as in aerospace and in automatic drug delivery, e.g., in closed-loop anesthesia.
The most ubiquitous controller is the proportional-integral-derivative (PID) controller used in approximately 90% of process control and mechatronics applications. A thorough coverage of the techniques available to design and tune PID controllers is available in the book by Åström and Hägglund,28 or in the article by Ang et al.29 The textbook version of the PID controller is given as
where e(t) is the error signal, u(t) the control signal, and the parameters K, T i, and T d are the tuning constants or gains. Figure 3 graphically depicts the 3 terms of the PID controller.
The selection of tuning constants constitutes the tuning process and is most critical for optimized operation. Although many empirical methods are available, they should be avoided as only systematic methods offer the necessary guarantees of stability and robustness. Depending on the control objectives, e.g., set point tracking, disturbance rejection, or robustness to uncertainty, different formal techniques are available to the engineer for tuning constant selection. When the controller is required to perform in the presence of significant uncertainty, tuning should guarantee stability (robust stability) and some minimal level of performance (robust performance) despite the uncertainty that is present. Given the fact that uncertainty is omnipresent, especially in closed-loop control of anesthesia, robust stability and performance are highly desirable characteristics of controllers.
Advanced Control Methodologies
Although simple, and very versatile, the PID controller is sometimes not appropriate, e.g., when the system to be controlled displays significant delay or when dealing with a multivariable system. A number of model-based control design techniques can then be used, resulting in controllers that are tailored to the specific application but are significantly more complex than the PID controller. One such class of controllers, very popular in the process industries, is model-predictive control, in which a model of the system is used to predict the future behavior of the system and the control signal is computed to minimize a criterion that involves future output errors and control signals. The minimization is usually solved as a constrained optimization problem.30 In some settings, a fixed robust controller is not sufficient to satisfy the specifications, in which case it may be necessary to use adaptive control, i.e., a controller that has the ability to automatically adjust its own tuning parameters to better fit the current patient. A technique combining both predictive and adaptive control known as extended predictive self-adaptive control has for instance been tested on simulated patients with promising results.31 A related technique called generalized predictive control has also been tested on 3 patients for control of muscular relaxation and unconsciousness during isoflurane anesthesia.32
For several decades now, part of the control community has been pursuing intelligent control, in which techniques from artificial intelligence are used in control systems. Techniques such as fuzzy logic, expert systems, neural networks, genetic algorithms, or reinforcement learning have been proposed, in particular for the control of processes that defy accurate mathematical description. An example in anesthesia is the use of Bayesian learning for tuning a BIS-guided control of propofol anesthesia that was tested on 20 patients.33 Moore et al.34 recently proposed reinforcement learning for the same problem and showed by means of simulations that this technique compares favorably with a manually tuned PID controller. For a recent review of intelligent control in anesthesia, see Lan et al.35
Many industrial control systems are deployed in a hierarchical manner, with master loops sending set points to slave loops. This is how some closed-loop control systems in anesthesia have been implemented, where a DOH controller sends a target concentration to a TCI system, which is itself an open-loop controller. Often the loop at the top is in charge of performing an economic optimization, as is often the case in industrial plants.36
Safety in Control
As in any engineered system, safety must be the primary concern. This has led to the development of a number of formal verification techniques to analyze the safety of control systems.37 More recently, techniques to synthesize closed-loop control systems that include a safety-preserving fallback mode have been developed for safety-critical systems.38 International standards for the development of physiologic closed-loop controllers (PCLCs) (International Electrotechnical Commission International Collateral Standard Requirements for the Development of Physiologic Closed-loop Controllers—IEC 60601-1-10:2007)39 have also been defined. This includes specification for the fallback mode and display of controlled variables as well as requirements for determining system characteristics such as standard control performance measures.
A number of steps are necessary for the rigorous development of an anesthesia control system before clinical trials (Table 1). These preliminary clinical trials should demonstrate the safety and efficiency of the system and precede the extended, ideally multicenter clinical studies, to demonstrate improved patient outcomes. Control algorithms developed by clinicians without the suggested formal control methodologies should be analyzed and tested with the assistance of control engineers. Serious flaws discovered should be remedied and fully tested in a preclinical evaluation. If clinical studies have already been performed with the flawed scheme, new clinical studies should be performed with the modified scheme. Ideally, to avoid such a scenario, control engineers should be involved from the start of the project, rather than as an afterthought.
CLINICAL STUDIES OF CLOSED-LOOP CONTROL OF ANESTHESIA
After ensuring a fast and safe induction, the anesthesiologist needs to maintain the patient in an adequate state of hypnosis, analgesia, and paralysis according to the requirements of the surgical procedure. The anesthetic and opioid titration need to be constantly adjusted to avoid both under- and overdosing of the patient. The idea of an automated system that would regulate drug dosing to maintain the adequacy of the anesthetic regimen would appear to be a logical solution. Closed-loop anesthesia would not replace anesthesiologists, but would allow them to concentrate on higher-level tasks.
A closed-loop controller for anesthesia should induce the patient rapidly, but with minimal overshoot and then maintain the patient in an adequate state of anesthesia and analgesia at least as well as an expert anesthesiologist. Translating this in control specifications is a challenging task, but can be attempted based on engineering principles, clinical input, and published data. For induction of anesthesia using an EEG-based index, it could be translated into a rise time (time to reach 90%–95% of the desired set point) at induction of 3 to 4 minutes, and an overshoot of less than 10% to 15%. During maintenance of anesthesia an EEG-based index, such as the BIS™ (Covidien, Mansfield, MA), should stay within 10 points of the target (typically between 40 and 60 for a BIS target of 50) 85% of the time. Indeed, most studies have considered this a target range for the BIS. Control specifications for analgesia should be such that in the case of arousal (which in control engineering terms can be thought of as an output disturbance), the patient response is rapidly suppressed, say within 2 minutes and without inducing oscillations (a characteristic of an unstable system). For IV anesthesia, the question of whether the closed-loop system should be based on TCI arises. There is no obvious engineering reason as to why TCI should be part of such a closed-loop system. However, in settings where TCI is widely accepted and used, a closed-loop system based on TCI may have better acceptance with clinicians. Conversely, in North America where TCI has yet to be approved by the Food and Drug Administration, such a choice would hinder the regulatory approval of closed-loop control. Another question that arises is whether one should develop a fully integrated system at once, or start with a simpler control system, e.g., for DOH and maybe analgesia, and then build on this foundation to add other aspects of anesthesia, e.g., hemodynamics. In aerospace, where high-fidelity models are available, the integrated, multivariable approach is standard. In anesthesia, due to the complexity of these systems and the lack of models describing all of the interactions, the step-by-step approach is the only practical solution. The appropriately designed system should drive improved clinical outcomes by improved hemodynamic stability, avoidance of excessively deep periods of anesthesia, faster emergence, and possibly reduced drug consumption. The main challenge is the inherent variability, both interpatient and intrapatient, thus robust stability and performance are of paramount importance.
The first efforts to automate anesthesia go back to the work of Mayo and Bickford in the early 1950s with their attempts to develop EEG-based automatic delivery of volatile anesthetics, see Bickford for an example.40 What follows is by no means an exhaustive review of the published work on closed-loop control of anesthesia. In the 1980s, a significant amount of work was performed on end-tidal concentration control for inhaled anesthetics such as halothane,41 closed-loop control of neuromuscular blockade,42 or mean arterial pressure control.43 In 1989, Schwilden et al.44 published the first work on closed-loop delivery of propofol guided by the EEG median frequency during a study on 11 healthy volunteers. For a review of the progress from 1949 to 1980, see the review by Chilcoat.45
The advent of the BIS™ monitor in the mid-1990s provided a commercial sensor that could be used as an input to a control system for hypnosis. This resulted in a significant increase in the number of both simulated and clinical studies on closed-loop control of anesthesia and particularly on controlling the depth of anesthesia.
One of the earliest efforts led by engineering groups was on the control of BIS during isoflurane anesthesia. This system was tested clinically on volunteers during minor surgical procedures.46 It consists of 2 cascaded loops, the slave loop being in charge of isoflurane end-tidal concentration that received its set point from the master loop in charge of maintaining the BIS value between 40 and 50. Both controllers were model-based internal model controllers. According to the authors, the system’s performance during these clinical tests was deemed “satisfactory,” indicating “higher-quality anesthesia for patients who were treated with automatic control,” without quantifying those statements. Anesthesia in these patients was manually induced with the controller taking over only for the maintenance phase. This group also considered control of MAP by closed-loop control of alfentanil using a PKPD model-based explicit predictive controller with constraints.47 Here too, “successful clinical tests” were performed on 13 patients. These 2 systems were combined for joint control of hypnosis and analgesia during clinical tests.48
Early efforts at BIS-guided closed-loop control of propofol infusion were performed by Absalom et al.49,50 who used a PID controller with empirical tuning to adjust the propofol effect-site concentration set point in a TCI system. Not surprisingly, the performance varied significantly from patient to patient with the system displaying oscillatory behavior characteristic of instability in some cases (3 of 10 cases in the first study). This system was revised and further tested on 20 patients, keeping the BIS within 15% of its set point of 50 for 79% of the time. All patients were induced using TCI mode, with an effect-site concentration chosen by the clinician who then switched to closed-loop control for the maintenance phase of anesthesia. A system with a similar structure is described by Liu et al.51 This controller is not strictly a PID controller, but a rule-based controller that is similar in principle to a PD controller. After empirical tuning, this system was tested against manual control of anesthesia in a randomized controlled trial involving 164 patients (83 in closed loop). The system was shown to outperform manual control in terms of BIS variability (89% vs 70% in the [40–60] target range), and resulted in similar hemodynamic stability. Puri et al.52 describe a heuristically tuned “adaptive” PID controller tested against manual control in a clinical trial involving 40 subjects, maintaining the BIS in the target range for 87% of the time, with similar hemodynamic stability as in manual control. A similar system is described in Hemmerling et al.53 in which what is reported to be a heuristic set of rules emulating a PD controller is tested against manual control in a clinical trial involving 40 subjects, with comparable results. Both studies unfortunately lack a detailed description of the control algorithms.
The systems previously described were designed heuristically and therefore their theoretical properties are difficult to assess. A rigorous approach to robust PID tuning for anesthesia is described in Dumont et al.54 where a PID controller is robustly tuned for a simulated population of 44 adults. Results of a feasibility study in adults using the NeuroSENSE DOH monitor (NeuroWave Systems Inc., Cleveland Heights, OH) showed that this simple controller provided clinically adequate anesthesia,55 with the control variable within the target range 88% of the time and a median time to induction of 4 minutes. This led to the development of a similar system for use in children. The results of a pilot study are reported in Soltesz et al.,56 van Heusden et al.,57 and West et al.,58 with the control variable within the target range 89% of the time and a median induction time of 3.8 minutes. Although preliminary results seem to indicate that a robust PID controller has clinically acceptable performance in the setting of significant interpatient uncertainty in a pediatric population, a patient-individualized scheme has been proposed and shown, so far in simulations only, to outperform a population-based controller.59 Sawaguchi et al.60 describe a model-predictive controller-based system for control of DOH that updates the patient-based model on induction data and is augmented by a set of rule-based fallback procedures. They tested their system on 79 patients and report excellent performance with the control variable within the target range 90% of the time. Reboso et al.61 describe a PI controller for BIS-guided DOH control. The PI controller is tuned on a simulated population of 12 patients, further fine-tuned on 10 real patients and then tested on 24 patients during maintenance control, after manual induction, with 83% of the time within the target range.
Liu et al.62 present what they term a dual loop that manipulates both the propofol and remifentanil infusion rates based on the BIS index alone. The basic concept is that a sudden increase in the BIS index is due to a nociceptive reaction and reflects an inadequate analgesic state. The controller that manipulates the remifentanil infusion rate combines a proportional action with a number of heuristic rules. Randomized clinical trials involving 167 patients during a wide variety of procedures showed that the system provides better control of the BIS index than manual control (82% vs 71% of the time within the target range), similar hemodynamic stability but accompanied by increased remifentanil consumption (0.22 vs 0.16 µg/kg/min). This system, like its predecessor, induces the patient using a TCI mode, with manually set targets for both propofol and remifentanil effect-site concentrations. The same group recently extended its control system using the M-Entropy monitor (GE-Healthcare, Helsinki, Finland) to the control of both the DOH as measured by the State Entropy while using the difference between the Response Entropy and the State Entropy as a measure of nociception.63 In a clinical study involving 61 subjects, results comparable with their BIS-based system were obtained, i.e., 80% of the time within the target range. Hemmerling and Charabti64 describe a system called McSleepy with control of DOH based on the BIS, analgesia based on his Analgoscore, and muscular relaxation using phonomyography. Although McSleepy is reported to provide adequate anesthesia, with the BIS within 20% of its target for 83% of the time, little technical detail and clinical results are publicly available. Janda et al.65 independently controlled both BIS-based DOH and muscular relaxation based on electromyography66 on 20 patients during maintenance of anesthesia. For DOH, a controller (based on fuzzy proportional, differential plus integral) attempts to maintain a target of 40 and is said to be “able to maintain the target values with a high level of precision in a clinical setting,” keeping the BIS within 10% of its target 55% of the time.
THE PATH TOWARD COMMERCIALIZATION
Manberg et al.67 provide an overview of the expected regulatory challenges facing closed-loop control of anesthesia. There are in addition a number of key engineering issues to be addressed for the successful development of joint anesthesia–analgesia closed-loop control systems. The aforementioned International Electrotechnical Commission standard specifies requirements for the development (analysis, design, verification, and validation) of a physiological closed-loop controller (PCLC). Most of the reported prototype systems have not been developed using structured medical device software development and design control processes, nor have they included robust performance characterization that would be required by regulatory authorities. To comply with those requirements, it is necessary to implement a design control process encompassing everything from the control theory behind the PCLC, the sensors and actuators, the hardware, the user interface, and the software. A rigorous design control process will have to be followed through the development process, for every component of the system, and every decision will have to be accurately documented. This precludes the use of Software of Unknown Pedigree. Mathematical proofs of stability and robustness must be developed and provided, for which the ability to mathematically characterize every component of the system will be required. Fallback and failsafe modes must be an integral part of the system, rather than an afterthought and a thorough risk analysis must also be performed. For every identified hazard, a mitigation strategy and the evidence to support claims of safety will have to be provided. Usability studies will need to be performed to evaluate human factors and the hazards that can result from human interaction with the system.
Only after approval by regulatory bodies should feasibility studies then be conducted. It is only once a prototype satisfying those requirements has been developed and tested for preliminary evidence of feasibility, effectiveness, and safety, that large-scale clinical trials, necessary to demonstrate the clinical outcomes of the PCLC, can be conducted. Indeed the main objective to justify widespread adoption of closed-loop control of anesthesia should be to demonstrate improved (or at least equivalent) outcomes. Clinical trials will thus be required to demonstrate not only the safety of the system, but should also clearly demonstrate tangible benefits to the patients. Given the current level of safety in anesthesia, these trials are likely to involve multiple centers and thousands of patients. A potential benefit of tighter anesthesia control would be a reduction in the occurrence of the so-called “triple-low” effect. Indeed, recently published evidence8 indicates that a cumulative time >15 minutes at a “triple low” of low MAP, low DOH (BIS), and low minimum alveolar concentration of volatile anesthesia may be associated with an extended hospital stay and increased mortality at 30 days. In addition, if the recent reports of a reduction in postoperative delirium and cognitive decline with anesthesia guided by processed EEG68,69 can be confirmed, significant outcome advantages could be realized by automation of drug delivery.
Finally, the question of intellectual property surrounding closed-loop control of anesthesia has to be resolved to attract potential investors. The concept of feedback itself is not patentable, and most control algorithms for anesthesia that have been published are based on known methodologies. What are more likely to be patentable are methods for tuning, ensuring safety, or the use of a particular sensor in the loop as preferred embodiment of the invention. Also, the process control industry tends to rely on protecting their solutions with undisclosed know-how and trade secrets, and control vendors may publish their basic control methodology to establish technical credibility.
The introduction of automation promises to reduce the variability and increase the safety of many processes in anesthesia, including automated anesthetic drug delivery. The ubiquitous real gains in performance promised by adoption of feedback control can be realized in anesthesia but will necessitate a strong engineering approach to the design, analysis, validation, and verification of closed-loop systems. Given that these elements are secured, and if major suppliers of anesthesia equipment are engaged in participating in the development and testing of such a system, we may in a few years see widespread use of closed-loop control of anesthesia and analgesia in daily clinical practice.
Name: Guy A. Dumont, PhD, PEng.
Contribution: Guy A. Dumont helped prepare this manuscript.
Attestation: Guy A. Dumont approved the final manuscript.
Conflicts of Interest: Guy A. Dumont is co-inventor of the NeuroSENSE monitor (NeuroWave Systems Inc., Cleveland, OH) and has consulted for NeuroWave Systems Inc. He has also consulted for GE Healthcare.
Name: J. Mark Ansermino, MBBCh, MSc (Inf), FFA (SA), FRCPC.
Contribution: J. Mark Ansermino helped prepare this manuscript.
Attestation: J. Mark Ansermino approved the final manuscript.
Conflicts of Interest: J. Mark Ansermino has consulted for GE Healthcare.
This manuscript was handled by: Dwayne R. Westenskow, PhD.
2. Khammash M, El-Samad H. Systems biology: from physiology to gene regulation. IEEE Control Systems Magazine. 2004;24:62–76
3. Iglesias PA, Khammash M, Munsky B, Sontag ED, Del Vecchio D. Systems biology and control—a tutorial. 46th IEEE Conference on Decision and Control. 2007;1:12
4. Rinehart J, Liu N, Alexander B, Cannesson M. Closed-loop systems in anesthesia: is there a potential for closed-loop fluid management and hemodynamic optimization? Anesth Analg. 2012;114:130–43
5. Kurz A, Sessler DI, Lenhardt R. Perioperative normothermia to reduce the incidence of surgical-wound infection and shorten hospitalization. Study of Wound Infection and Temperature Group. N Engl J Med. 1996;334:1209–15
6. Hamilton MA, Cecconi M, Rhodes A. A systematic review and meta-analysis on the use of preemptive hemodynamic intervention to improve postoperative outcomes in moderate and high-risk surgical patients. Anesth Analg. 2011;112:1392–402
7. Monk TG, Saini V, Weldon BC, Sigl JC. Anesthetic management and one-year mortality after noncardiac surgery. Anesth Analg. 2005;100:4–10
8. Sessler DI, Sigl JC, Kelley SD, Chamoun NG, Manberg PJ, Saager L, Kurz A, Greenwald S. Hospital stay and mortality are increased in patients having a “triple low” of low blood pressure, low bispectral index, and low minimum alveolar concentration of volatile anesthesia. Anesthesiology. 2012;116:1195–203
9. Watson PL, Shintani AK, Tyson R, Pandharipande PP, Pun BT, Ely EW. Presence of electroencephalogram burst suppression in sedated, critically ill patients is associated with increased mortality. Crit Care Med. 2008;36:3171–7
10. Leslie K, Myles PS, Forbes A, Chan MT. The effect of bispectral index monitoring on long-term survival in the B-aware trial. Anesth Analg. 2010;110:816–22
11. Lindholm ML, Träff S, Granath F, Greenwald SD, Ekbom A, Lennmarken C, Sandin RH. Mortality within 2 years after surgery in relation to low intraoperative bispectral index values and preexisting malignant disease. Anesth Analg. 2009;108:508–12
12. Kertai MD, Pal N, Palanca BJ, Lin N, Searleman SA, Zhang L, Burnside BA, Finkel KJ, Avidan MSB-Unaware Study Group. . Association of perioperative risk factors and cumulative duration of low bispectral index with intermediate-term mortality after cardiac surgery in the B-Unaware Trial. Anesthesiology. 2010;112:1116–27
13. Kertai MD, Palanca BJ, Pal N, Burnside BA, Zhang L, Sadiq F, Finkel KJ, Avidan MSB-Unaware Study Group. . Bispectral index monitoring, duration of bispectral index below 45, patient risk factors, and intermediate-term mortality after noncardiac surgery in the B-Unaware Trial. Anesthesiology. 2011;114:545–56
14. Cohen NH. Anesthetic depth is not (yet) a predictor of mortality! Anesth Analg. 2005;100:1–3
15. Mackey DC. Can we finally conquer the problem of medical quality? The systems-based opportunities of data registries and medical teamwork. Anesthesiology. 2012;117:225–6
16. Kheterpal S. Random clinical decisions: identifying variation in perioperative care. Anesthesiology. 2012;116:3–5
17. Liu N, Chazot T, Hamada S, Landais A, Boichut N, Dussaussoy C, Trillat B, Beydon L, Samain E, Sessler DI, Fischler M. Closed-loop coadministration of propofol and remifentanil guided by bispectral index: a randomized multicenter study. Anesth Analg. 2011;112:546–57
18. Albertos P, Mareels I Feedback and Control for Everyone. 2010 Berlin, Germany Springer Berlin– Heidelberg
19. Hjalmarsson H. From experiment design to closed-loop control. Automatica. 2005;41:393–438
20. Ljung L System Identification: Theory for the User. 19992nd ed Upper Saddle River, NJ Prentice Hall
21. Slotine JJE, Li W Applied Nonlinear Control. 1990 Upper Saddle River, NJ Prentice Hall
22. Absalom AR, Struys MMRF Overview of Target Controlled Infusions and Total Intravenous Anaesthesia. 2007 Gent, Belgium Academia Press
23. Guzmán JL, Hägglund T, Åström KJ, Dormido S, Berenguel M, Piguet Y Feedforward Control Concepts Through Interactive Tools. 2011 Milano, Italy 18th IFAC World Congress:6361–6
24. Mandel JE, Sarraf E. The variability of response to propofol is reduced when a clinical observation is incorporated in the control: a simulation study. Anesth Analg. 2012;114:1221–9
25. Black HS. Inventing the negative feedback amplifier. IEEE Spectrum. 1977;14:54–60
26. Åström KJ, Murray RM Feedback Systems: An Introduction for Scientists and Engineers. 2008 Princeton, NJ Princeton University Press
27. Zhou K, Doyle JC Essentials of Robust Control. 1997 Upper Saddle River, NJ Prentice Hall
28. Åström KJ, Hägglund T Advanced PID Control. 2006 Research Triangle Park, NC The Instrumentation, Systems, and Automation Society
29. Ang KH, Chong GCY, Li Y. PID control system analysis, design, and technology. IEEE Trans Control Syst Technol. 2005;13:559–76
30. Maciejowski JM Predictive Control with Constraints. 2002 Upper Saddle River, NJ Prentice Hall
31. Ionescu CM, De Keyser R, Torrico BC, De Smet T, Struys MM, Normey-Rico JE. Robust predictive control strategy applied for propofol dosing using BIS as a controlled variable during anesthesia. IEEE Trans Biomed Eng. 2008;55:2161–70
32. Mahfouf M, Linkens DA Generalized Predictive Control and Bioengineering. 1998 London, UK Taylor & Francis
33. De Smet T, Struys MM, Neckebroek MM, Van den Hauwe K, Bonte S, Mortier EP. The accuracy and clinical feasibility of a new bayesian-based closed-loop control system for propofol administration using the bispectral index as a controlled variable. Anesth Analg. 2008;107:1200–10
34. Moore BL, Quasny TM, Doufas AG. Reinforcement learning versus proportional-integral-derivative control of hypnosis in a simulated intraoperative patient. Anesth Analg. 2011;112:350–9
35. Lan JY, Abbod MF, Yeh RG, Fan SZ, Shieh JS. Review: intelligent modeling and control in anesthesia. J Med Biol Eng. 2012;32:293–308
36. Farmer TG Jr, Edgar TF, Peppas NA. The future of open- and closed-loop insulin delivery systems. J Pharm Pharmacol. 2008;60:1–13
37. Aubin JP, Bayen AM, Saint-Pierre P Viability Theory: New Directions. 2011 Berlin, Germany Springer Berlin–Heidelberg
38. Tomlin CJ, Lygeros J, Shankar Sastry S. A game theoretic approach to controller design for hybrid systems. Proc IEEE Inst Electr Electron Eng. 2000;88:949–70
39. International Electrotechnical Commission. Medical Electrical Equipment, Part 1–10: General Requirements for Basic Safety and Essential Performance—Collateral Standard: Process Requirements for the Development of Physiologic Closed-Loop Controllers. 2007 Geneva, Switzerland IEC
40. Bickford RG. Automatic electroencephalographic control of general anesthesia. Electroencephalog Clin Neurophysiol. 1950;2:93–6
41. Zbinden AM, Frei F, Westenskow DR, Thomson DA. Control of end-tidal halothane concentration. Part B: verification in dogs. Br J Anaesth. 1986;58:563–71
42. Brown BH, Perks R, Anthony M, Asbury J, Linkens DA. Closed-loop control of muscle relaxation during surgery. Clin Phys Physiol Meas. 1980;1:203–10
43. Monk CR, Millard RK, Hutton P, Prys-Roberts C. Automatic arterial pressure regulation using isoflurane: comparison with manual control. Br J Anaesth. 1989;63:22–30
44. Schwilden H, Stoeckel H, Schüttler J. Closed-loop feedback control of propofol anaesthesia by quantitative EEG analysis in humans. Br J Anaesth. 1989;62:290–6
45. Chilcoat RT. A review of the control of depth of anaesthesia. Trans Inst M C. 1980;2:38–45
46. Gentilini A, Frei CW, Glattfedler AH, Morari M, Sieber TJ, Wymann R, Schnider TW, Zbinden AM. Multitasked closed-loop control in anesthesia. IEEE Eng Med Biol Mag. 2001;20:39–53
47. Gentilini A, Schaniel C, Morari M, Bieniok C, Wymann R, Schnider T. A new paradigm for the closed-loop intraoperative administration of analgesics in humans. IEEE Trans Biomed Eng. 2002;49:289–99
48. Gentilini A, Rossoni-Gerosa M, Frei CW, Wymann R, Morari M, Zbinden AM, Schnider TW. Modeling and closed-loop control of hypnosis by means of bispectral index (BIS) with isoflurane. IEEE Trans Biomed Eng. 2001;48:874–89
49. Absalom AR, Sutcliffe N, Kenny GN. Closed-loop control of anesthesia using Bispectral index: performance assessment in patients undergoing major orthopedic surgery under combined general and regional anesthesia. Anesthesiology. 2002;96:67–73
50. Absalom AR, Kenny GN. Closed-loop control of propofol anaesthesia using bispectral index: performance assessment in patients receiving computer-controlled propofol and manually controlled remifentanil infusions for minor surgery. Br J Anaesth. 2003;90:737–41
51. Liu N, Chazot T, Genty A, Landais A, Restoux A, McGee K, Laloë PA, Trillat B, Barvais L, Fischler M. Titration of propofol for anesthetic induction and maintenance guided by the bispectral index: closed-loop versus manual control: a prospective, randomized, multicenter study. Anesthesiology. 2006;104:686–95
52. Puri GD, Kumar B, Aveek J. Closed-loop anaesthesia delivery system (CLADS) using bispectral index: a performance assessment study. Anaesth Intensive Care. 2007;35:357–62
53. Hemmerling TM, Charabati S, Zaouter C, Minardi C, Mathieu PA. A randomized controlled trial demonstrates that a novel closed-loop propofol system performs better hypnosis control than manual administration. Can J Anaesth. 2010;57:725–35
54. Dumont GA, Martinez A, Ansermino JM. Robust control of depth of anesthesia. Int J Adapt Control Signal Process. 2009;23:435–54
55. Dumont GA, Liu N, Petersen C, Chazot T, Fischler M, Ansermino JM. Closed-loop administration of propofol guided by the NeuroSENSE: clinical evaluation using robust proportional-integral-derivative design. Conf Proc American Society of Anesthesiologists (ASA). October 15–19, 2011 Chicago, IL:A1170
56. Soltesz K, van Heusden K, Dumont GA, Hägglund T, Petersen C, West N, Ansermino JM. Closed-loop anesthesia in children using a PID controller: a pilot study Conf Proc IFAC Conference on Advances in PID Control. March 28–30, 2012 Brescia, Italy
57. van Heusden K, Dumont GA, Soltesz K, Petersen C, West N, Ansermino JM. Clinical evaluation of closed-loop controlled propofol infusion in children Conf Proc World Congress of Anaesthesiologists. March 25–30, 2012 Buenos Aires, Argentina:Paper no. 616
58. West N, Dumont GA, van Heusden K, Khosravi S, Petersen C, Ansermino JM. The administration of closed-loop control of anesthesia for gastrointestinal endoscopic investigations in children Conf Proc Pediatric Anesthesiology. February 23–26, 2012 Tampa, FL:SP-241
59. Soltesz K, Hahn JO, Dumont GA, Ansermino JM. Individualized PID control of depth of anesthesia based on patient model identification during the induction phase of anesthesia Conf Proc IEEE Conference on Decision and Control and European Control Conference (CDC-ECC). December 12–15, 2011 Orlando, FL:855–60
60. Sawaguchi Y, Furutani E, Shirakami G, Araki M, Fukuda K. A model-predictive hypnosis control system under total intravenous anesthesia. IEEE Trans Biomed Eng. 2008;55:874–87
61. Reboso JA, Méndez JA, Reboso HJ, León AM. Design and implementation of a closed-loop control system for infusion of propofol guided by bispectral index (BIS). Acta Anaesthesiol Scand. 2012;56:1032–41
62. Liu N, Chazot T, Hamada S, Landais A, Boichut N, Dussaussoy C, Trillat B, Beydon L, Samain E, Sessler DI, Fischler M. Closed-loop coadministration of propofol and remifentanil guided by bispectral index: a randomized multicenter study. Anesth Analg. 2011;112:546–57
63. Liu N, Le Guen M, Benabbes-Lambert F, Chazot T, Trillat B, Sessler DI, Fischler M. Feasibility of closed-loop titration of propofol and remifentanil guided by the spectral M-Entropy monitor. Anesthesiology. 2012;116:286–95
64. Hemmerling T, Charabti S. McSleepy: a completely automatic anesthesia delivery system. Conf Proc American Society of Anesthesiologists (ASA). October 17–21, 2009 New Orleans, LA:A460
65. Janda M, Simanski O, Bajorat J, Pohl B, Noeldge-Schomburg GF, Hofmockel R. Clinical evaluation of a simultaneous closed-loop anaesthesia control system for depth of anaesthesia and neuromuscular blockade*. Anaesthesia. 2011;66:1112–20
66. Simanski O, Janda M, Schubert A, Bajorat J, Hofmockel R, Lampe B. Progress of automatic drug delivery in anaesthesia—the ‘Rostock assistant system for anaesthesia control (RAN)’. Int J Adapt Control Signal Process. 2009;23:504–21
67. Manberg PJ, Vozella CM, Kelley SD. Regulatory challenges facing closed-loop anesthetic drug infusion devices. Clin Pharmacol Ther. 2008;84:166–9
68. Chan MT, Cheng BC, Lee TM, Gin TThe CODA Trial Group. . BIS-guided anesthesia decreases postoperative delirium and cognitive decline. J Neurosurg Anesthesiol. 2013;25:33–42
69. Ballard C, Jones E, Gauge N, Aarsland D, Nilsen OB, Saxby BK, Lowery D, Corbett A, Wesnes K, Katsaiti E, Arden J, Amoako D, Amaoko D, Prophet N, Purushothaman B, Green D. Optimised anaesthesia to reduce post operative cognitive decline (POCD) in older patients undergoing elective surgery, a randomised controlled trial. PLoS One. 2012;7:e37410