TRY to imagine this: a dozen operating rooms filled with patients having surgery while a lone anesthesiologist, gazing at live data and video feeds on a tablet computer or wall of computer monitors, manages their general anesthetic from a distant location. Far fetched? If you think so, consider that earlier this year an IBM computer dubbed “Watson” beat the best human contestants on the television game show Jeopardy  , your iPhone can now talk back, and in some hospitals, the sickest patients are managed remotely by critical care physicians who may not be in the same time zone, let alone same building.1 Of course, there are many hurdles to clear (e.g.  , airway management and line placement) before this futuristic scenario could become reality for intraoperative anesthesia. Arguably one of the main ones is that only a trained observer using clinical judgment can determine the adequacy of unconsciousness and make appropriate adjustments in drug dosing to assure adequate surgical anesthesia. The article by Liu et al.  2 in this issue of the Journal provocatively challenges that assumption and moves us a step closer to the future.

“… the contest [on titrated administration of propofol and remifentanil] between man and machine was a tie. That itself is thought-provoking and potentially important.”

Liu et al.  2 tested the hypothesis that two machines—an infusion pump running software that calculates effect-site concentrations and a processed electroencephalographic device that analyzes a parameter called entropy—could outperform skilled manual control of drug titration by an experienced clinician during surgery using total intravenous anesthesia with propofol and remifentanil. In this context, entropy is essentially a measure of the disorder in the electroencephalogram signal; the electroencephalogram of an awake person will have high entropy, whereas an isoelectric electroencephalogram will have no entropy. What makes this randomized controlled trial interesting is that it took advantage of the fact that electroencephalogram entropy analysis generates measures of both hypnosis (i.e.  , state entropy [SE]) and analgesia (i.e.  , response entropy) and programmed the infusion controller to adjust the corresponding agent (propofol for hypnosis, remifentanil for analgesia) appropriately.

Note that this was not a test of electroencephalogram-guided drug delivery per se  ; both the infusion controller and clinicians had access to the same electroencephalogram data. Rather it was about who or what did a better job using the information. And use the information they did. During the maintenance phase, the dual-loop controller made nearly triple the number of dosing modifications per hour than did the clinicians (21 vs.  8 for propofol and 28 vs.  10 for remifentanil). Performance was rated on how well the machines and skilled clinicians did at keeping patients within predetermined electroencephalogram parameters (e.g.  , SE 40–60, defined as adequate anesthesia; deviation from target; intraindividual variability; and SE undershoot and overshoot), as well as by clinical measures such as drug consumption, hemodynamics, movement, time to tracheal extubation, and recall of intraoperative events. The machines bested the clinicians on some electroencephalogram measures of performance (e.g.  , Global Score [a combined measure of fraction of time the SE and response entropy were within the target range, performance error, and intraindividual variability]; fraction of time at “adequate anesthesia” as defined by SE 40–60; duration at excessive anesthesia as defined by an SE less than 40), but there were no differences between the groups on clinical endpoints.2 In essence, the machines did well what they were told (programmed) to do as far as managing electroencephalogram targets and matched the performance of skilled clinicians on clinical outcomes. Thus, the contest between man and machine was a tie. That itself is thought-provoking and potentially important.

The electroencephalogram has been the Holy Grail of anesthetic depth monitoring for more than half a century but has fallen on hard times lately, largely because the focus of the dialog changed from electroencephalogram as a monitor of “depth” to one of intraoperative awareness. As reviewed and commented upon recently3 – 6 consciousness and intraoperative awareness are neurobiologically exceedingly complex phenomena. This makes these states difficult to capture or evaluate with electroencephalography, no matter the parameter or sophistication of the processing algorithm. Recent studies examining the efficacy of the electroencephalogram bispectral index for minimizing the risk of intraoperative awareness confirm as much. Thus, although some controversy remains,7 two recent large prospective, randomized trials of high-risk patients found no benefit of bispectral index monitoring compared with the traditional practice of monitoring end-tidal anesthetic gas concentration.8,9 However, generic monitoring of anesthetic depth is something else entirely. The very term “depth” as it applies to consciousness and anesthesia is a nebulous nineteenth century descriptor that lacks scientific exactness. It is crude but yet clinically useful, just the sort of thing that a processed electroencephalogram probably can handle. After all, although limited in its ability to provide information about complex, integrated forebrain functions such as memory, the electroencephalogram is good at identifying whether a brain is active or inactive. As such, it may provide a window into brain activity or “depth” during general anesthesia that may be clinically useful. And that is the key point of the study by Liu et al.  2 Machine-driven, electroencephalogram-guided drug delivery, even to somewhat arbitrary endpoints of depth such as SE, response entropy, and time in burst suppression, produced an anesthetic that clinically was indistinguishable from that provided by an experienced human. After a series of recent defeats, score a win for processed electroencephalogram as a tool for titrating anesthetic depth.

But can a processed electroencephalogram and a drug pump do our job? Not likely, at least not yet. This was a small study (30 and 31 patients per group), the electroencephalogram target defined as adequate anesthesia (SE 40–60) is debatable and seemingly not adjusted for age, and the study was grossly underpowered to detect differences in the risk of intraoperative awareness. Validation in a larger and more diverse cohort is clearly necessary. Indeed, because of wide interindividual variation in the bispectral index or spectral entropy during administration of commonly used anesthetic agents, others have questioned the ability of these indices to differentiate reliably consciousness from unconsciousness.10 What's more, as the author's acknowledge, the measure used to guide administration of the analgesic remifentanil (i.e.  , response entropy) reflects activity of the frontalis muscle, which is not well validated as an index of nociception, pain, or sensibility. In fact, more complicated neural events and responses, including activation of cortical circuits, occurs in vegetative patients and is not necessarily a sign of self-awareness.11 Another important limitation is that the study used only two of the many medications commonly used clinically for general anesthesia. Most such medications, but especially ketamine, nitrous oxide, and dexmedetomidine, have electroencephalogram signatures different from those of propofol and remifentanil, and partly because of persistent disorder in the signal, it has proven difficult to find reliable electroencephalogram indices of unresponsiveness when these agents are used.12,13 Therefore, it is premature and perilous to conclude from the data of Liu et al.  2 that the electroencephalogram is a generally or broadly useful tool for guiding anesthetic “depth.” In addition, because no group was managed blind to the electroencephalogram, Liu et al.  2 have not demonstrated that electroencephalogram data add value beyond that provided by simply having an experienced clinician manage the anesthetic.

Still, as one of us suggested elsewhere,6 having a window into the brain during general anesthesia makes sense. It is also long overdue; 165 yr after the first public demonstration of ether anesthesia, we still rely on the patient to tell us whether our medications are having the desired effect on the central nervous system (e.g.  , “can you open your eyes?”), and then we get only yes-no, on-off information. The brain is more nuanced than that, and in 2012, we ought to be able to do better. The work of Liu et al.  2 suggests that the processed electroencephalogram can be such a window if the conditions are tightly controlled and the expectations are low. That is, if we ask the electroencephalogram to evaluate “depth,” not awareness, it can with little or no human intervention and a well-programmed infusor-controller do a fine job of guiding delivery of some drugs to produce acceptable clinical anesthesia and sedation. Whether or not it is justified and safe, it is easy to see such systems taking root in endoscopy suites and critical care units, where sedative or anesthetic medications are administered by nonanesthesia specialists for brief, relatively unstimulating procedures or for long periods under relatively stable conditions. Whether the quality of surgical anesthesia benefits from electroencephalogram guidance of “depth” is another matter and one not addressed by Liu et al.  2 The current processed electroencephalogram monitors are no panacea for anesthetic depth monitoring and likely will never be able to handle all eventualities, patients, and drug combinations. But considering how primitive the tools and how low the bar on “depth” assessment now, the foggy window into the brain the processed electroencephalogram provides might be enough to light the path forward.