There is mounting evidence that the common pathway leading to delirium is an activation of microglia cells—cerebral immune cells with the capacity to launch local reactions—with subsequent neuroinflammation [4, 60, 61].

Peripheral inflammation can have a profound influence on the brain through the dissemination of cytokines—namely, interleukin 1β, interleukin 6, and tumor necrosis factor-α (TNF-α) [61]. These proinflammatory cytokines are released by trauma, surgery, or infection, initiating a systemic response that also activates microglia in the central nervous system (Fig. 40.4). It is important to note that this communication is not limited to humoral processes, but can also occur directly via afferent neural pathways. Microglial cells are extremely sensitive to a variety of coexisting factors, so that a previous insult can prime these cells and a relatively mild subsequent insult could trigger an exponential reaction [4].

Willard et al. showed that by administering a peripheral injection of lipopolysaccharide in rats, an acute and chronic neuroinflammation could be triggered [62]. The levels of TNF-α, which has an established role in microglial activation, rose considerably in the periphery and in the brain, whereas the levels in the brain remained elevated for months thereafter [63]. The effects of a neuroinflammation through cytotoxic agents are not only acute, but can also persist due to structural damage to synapses and neuronal apoptosis. The chronic inflammation in Willard’s experiment induced a time-dependent, and not dose-dependent, neuronal loss of both choline acetyltransferase (responsible for acetylcholine synthesis) and p75-immunoreactive cells (responsible for the inhibition of apoptosis).

The cholinergic anti-inflammatory pathway, as presented by Tracey et al., showed that cholinergic inhibition of this inflammatory processes is key for limiting the extent of the reaction, thus avoiding an exaggerated response with excessive inflammation [61]. The role of this cholinergic inhibition has also been well established: it has been shown that stimulation of the vagus nerve suppresses inflammatory response, that a vagotomy exacerbates cytokine release [64], and that microglia itself is deactivated in the presence of the parasympathetic neurotransmitter acetylcholine [65]. This cholinergic inhibition can be hampered by a variety of factors, such as through medication (e.g., anticholinergic drugs, benzodiazepines), preexisting conditions (e.g., dementia, substance withdrawal), previous inflammation (prior structural damage, priming of microglia), or simply old age—predisposing the brain for delirium.

As postulated by van Gool, an additional insult to an already predisposed brain allows microglia—now unchecked by the cholinergic pathway—to become abnormally active, releasing cytokines that activate further microglia, thus entering a vicious cycle by triggering a sustained local inflammation with subsequent neurodegeneration, with further damage to cholinergic pathways [4]. This uncontrolled neuroinflammation, with neurochemical and synaptic disturbances, can explain the behavioral effects, as well as short- and long-term consequences of delirium and POCD. Additionally, this provides a plausible explanation as to the roles of many recognized risk factors, such as advanced age and the use of anticholinergic drugs, in the genesis of delirium.

Metabolic disorders also appear to play a significant role in the pathophysiology of delirium. Aging, neurological maladies, as well as diabetes and hyperglycemia seem to predispose the development of cognitive dysfunction.

The involvement of diabetes is not surprising, as this condition is known to induce vascular, sensory, and cognitive complications. Hyperglycemia is also known to affect a wide range of structures, such as the blood-brain barrier and synaptic connections, as well as directly increase the release of cytokines [66]. Coupled with neurotoxicity and impaired circulation, the scope of proinflammatory properties of diabetes becomes evident. A perioperative tight glycemic control has also been shown to have protective effects against POD/POCD.

Indirect effects of anesthesia, such as sedation and neurotoxicity, must also be considered in the genesis of delirium.

Much like the sedation-related delirium, which is known in the ICU context as being rapidly reversible, every hypnotic agent, or agent with sedative side effects, interacts with GABA and NMDA receptors. Interaction with those receptors leads to an inhibition of neuronal activity, affecting attention, qualitative and quantitative consciousness, as well as cognition [67]. By intermittently setting perfusion pump on and off, an acute onset and a fluctuating level of attention and/or consciousness over the day are easily produced, finally fulfilling all DSM-5 criteria for delirium. This form of delirium is a reaction to the termination of sedation and thus also related to emergence delirium.

Preclinical experimental work raised growing concerns regarding cognitive and behavioral impairments due to anesthesia. Several experimental trials established a link between time in anesthesia and dose-dependent calcium dysregulation and neuroapoptosis in growing mice brains [68, 69]. This effect has also been shown for surgery, surgical stress, and neuroapoptosis [70]. The significance of these animal studies for humans is still unclear, and there are ongoing prospective clinical trials aiming to clarify this issue.

The role of anesthesia must also be considered in the context of delirium pathophysiology. Extended periods in deep anesthesia, as expressed in a burst suppression pattern and duration in EEG monitoring, are also associated with postoperative delirium [71–74]. Burst suppression pattern represents a massive reduction of central activity and neuronal metabolic rate. Deep anesthesia may cause disturbance of neuronal homeostasis with the detrimental complication of POD. Studies published by Monk et al. show that cumulative time in burst suppression in noncardiac surgery patients significantly increased mortality within a 1-year period [75]. These results advocate that the use of EEG for the monitoring of anesthesia depth should be employed routinely, but especially when dealing with more vulnerable populations, such as the infant and elderly patients [76].

The individual risk of POD is determined by predisposing and precipitating risk factors, as suggested in a risk model that has been established in the late 1990s [50].

Predisposing factors are generally preexisting conditions that place the patient at an increased risk for the development of delirium. There are numerous predisposing risk factors that have been identified in the surgical context, including cognitive impairment, diabetes, anemia, history of stroke, previous delirium, as well as advanced age (Fig. 40.5).

Precipitating factors are triggers for delirium in a specific treatment framework, developing in the context of the medical treatment. Though these may be modifiable under certain circumstances, they are not always avoidable. Precipitating factors include, for example, the use of drugs with anticholinergic activity, burst suppression rate and duration under anesthesia, a prolonged period of fluid fasting, as well as poorly managed postoperative pain (Fig. 40.6).

A special consideration must be given to advanced age, as it is—from the quantitative point—an important and frequently reported risk factor [2, 3, 17, 37, 50, 77, 78]. However, current evidence suggests that chronological age should not be viewed strictly as a risk factor, but rather as a surrogate marker for comorbidity, multimorbidity, and a loss of functional reserve [79, 80]. Undoubtedly, both comorbidity and functional impairment are more often present with advanced age, but these factors are surely not limited to elderly patients. There is a high heterogeneity among the older population, which requires the inclusion of a detailed assessment in order to properly estimate the overall risk [81]. The reduction of the “functional cognitive reserve ” is the one of the most important factors to be considered in elderly patients. This means that fewer or less severe precipitating factors might suffice to induce delirium in those patients.

This increased vulnerability is not exclusively related to comorbidities like dementia (which can indeed occur at any age), but rather the loss of physiological functions. The functional status includes several domains, such as cognition, sensory functions, mobility, and malnutrition [49]. “Frailty ” indicates a severely impaired functional status that is not limited to one organ system, but rather denotes a systemic condition. It seems critical to understand that not all elderly patients are frail, but only a fraction (normally between 5 and 30 % in the general population) [82]. In the in-hospital surgical setting, it is estimated that about half of the elderly patients suffer from frailty (Fig. 40.7) [78, 80].

Finally, the role of age as a predisposition condition is caused by the higher likelihood of an accumulation of age-related risk factors, linked to comorbidity and functional impairment. Therefore, a detailed functional assessment in elderly patients is of utmost importance. This includes functional tests focusing on mobility and coordination, such as the “timed up and go” test; a cognitive screening with validated tools, such as the Minimal Mental State Examination (MMSE) ; and a detailed medical history accounting for malnutrition, sensory and hearing loss, as well as psychiatric disorders and comorbidities (Table 40.5) [83].

In clinical routine, it would be desirable to use validated tools that predict the actual risk for the development of POD for each individual patient. There are several risk prediction models that have been developed in different contexts. A recent systematic review and meta-analysis found 37 risk prediction models for POD, although the authors found only seven were either internally or externally validated [84].

While those models might indeed be more suitable for risk evaluation than an individual clinical decision [85], currently there is still no risk prediction model that adequately covers all patients. While a comprehensive risk model remains unavailable, the risk assessment should continue to be evaluated on an individual basis in the clinical setting (Fig. 40.8).

Postoperative delirium was first attributed to anesthesia rather than to the surgical procedure itself. In their article from 1955, Bedfort and Leeds attributed the risk of a patient developing delirium solely to the general anesthesia [1]. In the later phase of POD research, it became evident that it is rather the inflammatory stress induced by a surgical procedure that causes POD. Nevertheless, anesthesia and anesthesia-related factors are important aspects that can either place patients at risk or exert protective effects. The most important factors in this context are the type of anesthesia, hemodynamic management , neuromonitoring, and postoperative pain management [71, 86–88].

It is of interest to note that elderly patients are more likely to experience burst suppression during anesthesia [107]. Age-related changes during anesthesia with propofol and sevoflurane are related to EEG power and coherence. EEG power is defined as a function of EEG wave amplitudes and frequencies, whereas coherence can be seen as a frequency-dependent correlation or as a measure of synchrony between two signals at the same frequency (e.g., alpha-band) in different regions of the brain. Purdon and colleagues [107] examined 155 patients, aged 18–90 years old, while receiving either propofol or sevoflurane anesthesia. For both anesthetics, they found a marked reduction in EEG signal power for all frequency bands (a = 8–12 Hz, b = 13–35 Hz, q = 4–7 Hz, d < 4 Hz, g > 35 Hz) with increasing age. The effect was most pronounced in the alpha-frequency band, where they found a loss of coherence, as well as a lower peak coherent frequency in elderly patients, as compared to the younger patients . They proposed that the age-related EEG power reduction might be caused by a decline in synaptic density, changes in dendritic dynamics, or a decline in neurotransmitter synthesis within the cortex. The frontal alpha-band changes are thought to be mediated through the frontal GABA (gamma-aminobutyric acid) thalamocortical circuits [108], so that these age-related changes might reflect a functional alteration in the GABA-dependent fronto-thalamocortical circuits. In the propofol group, these occurrences were more pronounced, which can be related to the different underlying molecular mechanisms of both drugs: while propofol acts primarily as an activator on the GABA receptor, sevoflurane and other inhalative anesthetics show an additional inhibition on NMDA receptors [109].

Propofol binds postsynaptically to GABA receptors, hyperpolarizing postsynaptic neurons and thus leading to inhibition [109]. EEG signatures of propofol-induced loss of consciousness show an increase in low-frequency power, a loss of coherent occipital coherent alpha-oscillation, and the appearance of coherent frontal alpha-oscillation, which is reversed during regain of consciousness [110]. Additionally, there is a disruption of frontoparietal feedback connectivity, which also recovers during regain of consciousness (Fig. 40.9) [111].

Dexmedetomidine is an alpha-2-adrenoceptor agonist that gives rise to similarly slow oscillations and spindle-like activity during sedation [112]. In contrast to propofol, dexmedetomidine is clinically known to induce a sedation state comparable to non-rapid eye movement sleep, in which patients can easily be aroused with verbal or tactile stimuli. Both anesthetics are associated with slow/delta oscillation during induced unconsciousness, even though the amplitude power of slow wave oscillation was much larger in propofol anesthesia. Similar to sleep-induced spindles, unconsciousness induced by dexmedetomidine triggers spindles with a maximum power and coherence at 13 Hz, in contrast to propofol, where the peak frequency is at 11 Hz [97]. The authors propose that propofol enables a deeper state of unconsciousness , as seen by large-amplitude slow wave oscillation, whereas dexmedetomidine places patients into a more plane brain state of sedation.

Ketamine acts primarily via the inhibition of NMDA receptors, inducing a “dissociative anesthesia” [109]. This difference in molecular interaction can be seen in EEG analysis, where a reduction in alpha-power as well as an increase in gamma-power can be noted. But despite the molecular and neurophysiological differences to the other major classes of anesthetics, the frontoparietal feedback connectivity was gradually diminished during induction with ketamine anesthesia and was inhibited after loss of consciousness [113].

Target-controlled infusion (TCI) systems have been developed for intravenous drugs, where a set of pharmacokinetic parameters has been selected for computer simulation of an infusion scheme. The selected model is incorporated into the infusion pump, where it is used to predict the drug concentration in the plasma and at the drug target site. This allows it to calculate the needed concentration of anesthetics to reach an unconscious state. Monitoring the depth of anesthesia, or unconscious state during anesthesia, may be defined as the probability of nonresponse to stimulation [114] and, therefore, is dependent on the intensity of the stimulus, as well as the nature of response. Anesthesia may well induce unresponsiveness and amnesia, but the extent to which it causes unconsciousness remains uncertain [109].

Therefore, it may be more reliable in the future to focus on parameters that are directly related to the level of consciousness, such as the EEG, in order to monitor depth of anesthesia/unconscious state. In order to ensure more reliable results, it is important to use EEG signatures that are specific to age and to the anesthetic of choice.

To measure “arousability,” which is the balance between noxious stimulation and nociceptive suppression by analgesics, it is necessary to analyze responses evoked by a strong painful stimulus. Since analgesics primarily act at a subcortical level, it seems appropriate to assess electrophysiological reflexes at the subcortical/spinal level to achieve this goal.

In the last decades, the availability of short-acting drugs with a high degree of performance prediction, such as propofol [115], allowed for the development of novel approaches to anesthesia. Aside from total intravenous anesthesia (TIVA) , these advancements allowed for the development of target-controlled infusion (TCI) systems , which employ multi-compartment pharmacokinetic models to predict anesthetic doses [116, 117], as described in Chaps. 6 and 8.

Since the first computer-based infusions were developed in the 1980s [118, 119], increasingly more accurate and reliable models have expanded the prospects of TCI, and today there are regimens available for the drug delivery of several substances, such as sedatives, analgesics, antiarrhythmics, antibiotics, and chemotherapeutics (see Chap. 8).

Anesthesia is attained by a balanced mixture use of hypnotics and analgesics, whereas the ideal dosage is generally estimated using vegetative signs as surrogate markers, such as heart rate and blood pressure fluctuations. Monitoring the depth of anesthesia in this fashion is challenging and has several limitations, so that much skill is needed to properly recognize and interpret stress signals in a broad patient collective. While an insufficient depth of anesthesia increases the risk of awareness and subsequent complications [93, 120], excessive anesthesia, expressed as burst suppression patterns in EEG analysis , is associated with POD [72, 100]. Therefore, inadequate dosage of these substances in either side of the target range can have severe effects on patient outcome [40]. There are promising new monitoring approaches, such as the noxious stimulation response index (NSRI) [121] and surgical pleth index (SPI) [122, 123], which may be helpful devices in future anesthetic assessments.

By properly defining target doses, TCI offers an elegant solution to this predicament, with the potential to decrease complication rates and improve patient outcome. One major limitation of the most commonly used TCI algorithms is that anesthesiologists utilize algorithms that were designed for the use of one opioid (mostly remifentanil) and one hypnotic (mostly propofol). Usually, the target plasma concentrations do not allow an automatic correction for co-analgesics and co-hypnotics , so that these have to be manually adjusted. Especially for patients with a high risk for delirium, these co-substances might play an important role by blocking proinflammatory pathways, thus providing beneficial effects for patients. This can explain the protective effect of iv agents (e.g., ketamine, dexmedetomidine) and also their sparing effect on hypnotics with a considerable risk for burst suppression (e.g., propofol).

When using a TCI model in a patient with a significant risk for POD , the additional use of an EEG-controlled monitoring is recommended in order to avoid excessive anesthesia and to better titrate analgesia to avoid pain and overdosing of opioids by reduction of anticholinergic side effects, as well as to allow for manual adjustments of the target concentrations, if necessary.

Propofol is a highly lipid-soluble substrate that readily permeates biomembranes such as the blood-brain barrier, so that the onset of anesthetic effects is equivalent to the blood circulation time. This property also allows for the rapid redistribution to the periphery, so that patients can readily recover from anesthetic states.