Preclinical and translational models for delirium: Recommendations for future research from the NIDUS delirium network

Abstract

Delirium is a common, morbid, and costly syndrome that is closely linked to Alzheimer’s disease (AD) and AD-related dementias (ADRD) as a risk factor and outcome. Human studies of delirium have advanced our knowledge of delirium incidence and prevalence, risk factors, biomarkers, outcomes, prevention, and management. However, understanding of delirium neurobiology remains limited. Preclinical and translational models for delirium, while challenging to develop, could advance our knowledge of delirium neurobiology and inform the development of new prevention and treatment approaches. We discuss the use of preclinical and translational animal models in delirium, focusing on (1) a review of current animal models, (2) challenges and strategies for replicating elements of human delirium in animals, and (3) the utility of biofluid, neurophysiology, and neuroimaging translational markers in animals. We conclude with recommendations for the development and validation of preclinical and translational models for delirium, with the goal of advancing awareness in this important field.

1 |. INTRODUCTION

What Is Delirium?

Delirium is an acute confusional state that is common, morbid, and costly. It is characterized by both an acute-onset and fluctuating nature that involves cognitive disturbances (eg, inattention, an altered level of consciousness, and disordered thinking) and noncognitive features (eg, altered sleep-wake cycles). Primary features include (1) a reduced awareness of the surrounding environment (eg, difficulty focusing and/or often changing topics; being easily distracted; withdrawal or absence of response to surroundings), (2) poor thinking skills (eg, not knowing where or who others are, forgetting recent events, rambling or nonsense speech), and (3) behavioral and emotional changes (eg, anxiety, fear or distrust; restlessness; being quiet or withdrawn). Symptoms of delirium typically begin suddenly (eg, within a few hours to days) and can often come and go throughout the day. Moreover, signs of delirium often worsen in unfamiliar settings (eg, hospital, intensive care unit [ICU], nursing home).

Delirium is currently characterized by three major subtypes: (1) the more obvious hyperactive type, (2) the frequently missed, though more common, hypoactive type, and (3) a combination of hyperactive and hypoactive types seen in mixed delirium. Hyperactive delirium includes behaviors that range from restlessness to agitation and constant movement.^1,2 Hypoactive delirium is characterized by at least one of the following: paucity of speech with or without prompting, slow or no movement, or unresponsiveness.¹ Mixed delirium can present as switching back and forth between restlessness (a hyperactive feature) and sluggishness (a hypoactive feature). This broad phenomenological range, coupled with the fluctuating course of delirium, makes this syndrome particularly challenging to identify and determine its underlying pathophysiological mechanisms.

Delirium occurs in ≈20% of older patients admitted to a general medicine ward, >20% of high-risk patients following major surgery, and 50% to 70% of mechanically ventilated patients in critical care settings.³ Delirium is not only emotionally disturbing to patients, families, and caregivers, but it is strongly associated with many adverse outcomes, including longer length of hospital stay, greater morbidity and mortality, higher rates of institutionalization, and increased risk of long-term cognitive decline and dementia.^4–7 Given its association with increased health care utilization and costs, delirium holds substantial implications for public health and health policy. For example, developing delirium during an intensive care unit (ICU) stay was associated with a 39% increase in the cost of ICU care. More recent research has shown 30-day cumulative incremental costs attributable to ICU delirium upwards of $22,000 per patient. It is estimated that total direct 1-year U.S. health care costs attributable to delirium range from $143 to $152 billion for all older persons and from $26 to $42 billion per year in older surgical patients,⁸ costs rivaling those associated with cancer and diabetes.

Given the associated poor outcomes and increased costs attributed to delirium, the imperative for delirium mitigation is pronounced. With empirical studies suggesting the partial effectiveness of nonpharmacological delirium mitigation strategies and given the established epidemiological link between delirium and dementia (eg, potential acceleration of dementia following an episode of delirium), delirium may be a preventable contributor to the increasing burden of dementia. This hypothesis has been highlighted via recent calls for a public health campaign to mitigate delirium as a means of reducing the global burden of dementia.⁹

What Is Known about the Biological Basis of Delirium?

In recent decades, clinical studies of delirium have advanced our knowledge in terms of its incidence and prevalence, risk factors, biomarkers, outcomes, prevention, and management, topics that have been comprehensively reviewed in recent work.^3,7,10–13 However, the fundamental understanding of delirium neurobiology remains unclear. Several well-studied areas showing promising associations with delirium include neuroinflammation, neuronal injury, and neurotransmitter imbalance.^3,14–23 To more fully understand the complex neurobiological underpinnings of delirium, it will be necessary to identify causal mechanisms. To determine such causal mechanisms, experimental studies in animal models provide the most common and compelling approach. Although this is a developing area, high-quality science in this area will help advance our understanding of the fundamental neurobiology of this clinical syndrome. Table 1 presents key conceptual goals for the use of animal models in delirium research and includes some practical examples of experimental questions.

TABLE 1.

Uses of animal models in delirium research

Conceptual goal	Practical examples of experimental questions
Advance pathophysiological and mechanistic understanding of delirium Apply experimental manipulations to test mechanistic hypotheses Determine causal mechanisms and pathways Develop, test, and refine specific interventions for prevention and treatment	Does exercise preconditioning reduce incidence/severity of delirium in animal models and through which mechanistic pathways?
Validate and extend clinical findings Directly link mechanisms to pathophysiological, phenotypic, or neuroanatomic abnormalities	Is slow EEG activity causally associated with inflammation in an animal model ofdelirium? Is inflammation-related EEG slowing mechanistically associated with a phenotype of behavioral quiescence?
Serve as a platform for validation of human biomarkers of delirium	Do proteins identified from proteomic approaches in human samples associate with rodent behavioral tests align with human delirium features?

Abbreviation: EEG, electroencephalogram.

1.1 |. Benefits of using animal models in delirium research

Multifactorial contributors to delirium have been identified in human studies, including both vulnerability and precipitating factors.¹⁰ Animal models allow investigators to carefully consider these individual factors to rigorously test mechanistic hypotheses and develop potential interventions and treatments. Notably, animal models have been used extensively for preclinical testing of many human treatments, such as for Alzheimer’s disease (AD) and AD-related dementias (AD/ADRD),¹¹ Parkinson’s disease,¹⁴ multiple sclerosis,¹⁵ and heart failure.¹⁶ Although progress in these areas has been slow and plagued with setbacks, it is relevant to note that the only currently approved disease-modifying therapy for AD/ADRD has been based on decades of animal model research.¹⁷ Importantly, since AD/ADRD and mild cognitive impairment (MCI) are strong risk factors for human delirium, the availability of multiple rodent models of different aspects of these neurodegenerative disorders could be leveraged to enhance the validation of studies focused on delirium phenotypes.

The use of animal models to observe disease-associated motor neuron degeneration, for which brain tissue in humans is rarely available, has informed the development of therapeutic interventions for amyotrophic lateral sclerosis.¹⁸ In this paper, we adopt the definition of a potential biomarker given by the National Institutes of Health (NIH) Biomarkers Definitions Working Group panel: a “characteristic that is objectively measured and evaluated as an indicator of normal biological processes, pathogenic processes, or pharmacological responses to a therapeutic intervention.”²⁴ Here we use the term biomarker as an indicator of a pathogenic processes pertaining to delirium.

To guide the future development of animal models for delirium and enhance understanding of delirium neurobiology, the Network for the Investigation of Delirium: Unifying Scientists (NIDUS) assembled a group of scientific investigators actively working in the field of delirium and/or animal models with expert knowledge from varied settings in which delirium occurs (eg, medical, surgical, and ICU). A major goal of the NIH-funded NIDUS is to identify research priorities for the field.²⁵ Such studies are conducted on a voluntary basis and have followed a standard format in identifying experts across disciplines, proposing areas to be addressed, forming writing groups, reviewing relevant literature, and drafting summaries and recommendations. During 2020, the development of animal models for delirium was identified by NIDUS as a research priority for the field.²⁵ To advance a proposed paper, a core group of NIDUS members (SMV, NL, JN, GC, SKI, NT, RGE) met virtually five times from October 2020 to March 2021 (October 30, November 4, and December 15, 2020, and February 2 and March 2, 2021) to discuss (1) the current state of knowledge of animal models of delirium (JN, GC, JAH), (2) the development of animal models of delirium, including general principles and realistic goals (TA, SB, RGE), (3) phenotypic and behavioral assessments of delirium features in animal models (NL, LD, YI, NT), and (4) approaches to validating preclinical models of delirium (TA, SB, RGE). Subgroups were formed to address each of these four topics and met separately for one to three sessions each to further develop and refine each of these points (individual membership of each of the aforementioned subgroups). The subsections were combined into a single document, with introductory sections and tables drafted as needed by the core group. Additional members provided critical feedback on drafts of the manuscript (LA, CC, RdC, SL, ERM, ZX). Consensus was achieved through an iterative feedback process with all members of the group.

It is important to acknowledge that we are at an early stage in the development of animal model research in delirium. We had hoped that this expert consensus paper might provide preliminary recommendations for the field to move forward and to incorporate best practices and lessons learned from other fields, such as AD and aging. Since our goal was to provide preliminary research recommendations for a developing field and not to develop more advanced guidelines at this early stage, elements of the Grading of Recommendations, Assessment, Development and Evaluations (GRADE) process, such as input from pharmaceutical industry personnel, funding agencies, laypersons, and patients, were considered outside the scope of this project.

1.2 |. Validation of preclinical models of delirium

A recurring theme in this work is that no animal model perfectly recapitulates all aspects of a human condition, especially a complex behavioral syndrome like delirium, but that validity needs to be considered in the context of the specific experimental goal. Common model validity frameworks include face validity, construct validity, predictive validity, and target validity²⁶ (see Table 2 for applications to animal models). For example, a delirium model would have high face validity if the animal demonstrated a recognizable delirium-like phenotype (detailed in what follows) and high construct validity if this occurred in a context associated with delirium in humans and involving suspected elements of human delirium pathophysiology. Examples of delirium models that combine reasonable face and construct validity are sepsis, general anesthesia, and anesthesia/surgery models. Other elements of construct validity are not yet available for delirium. There are currently no well-established genetic predispositions (eg, candidate genetic markers) that could be adapted to create genetic models for delirium, as has been productively utilized for AD/ADRD. Although there are no definitive biomarkers for delirium, several promising markers have been identified (see Biomarkers section in what follows). In addition, there are no definitive objective measures of delirium in rodents, although this is an area of highly active investigation.

TABLE 2.

Criteria for assessing validity of animal models

Validity type	Criteria for assessment
Face validity	Model should mimic aspects of the clinical presentation (see Table 4 for specific examples of behavioral tests relevant to delirium in rodents)
Construct validity	Model should produce signs of the condition by mimicking the types of etiological situations that lead to this symptomology
Predictive validity	Model should respond to specific treatments in a manner similar to patients that experience the disorder
Target validity	Model should demonstrate impact on specific pathophysiological pathways known to be associated with the condition

Predictive validity is the ability to predict outcomes related to delirium, such as clinical outcomes or the response to targeted interventions. Rodent models of depression may have uncertain face validity, but the response to human antidepressants provides a useful degree of predictive validity.²⁷ Delirium currently lacks pharmacological therapies, although as mentioned earlier, nonpharmacological interventions to prevent delirium may provide an as-yet underexplored means to test the predictive validity of delirium animal models. Finally, target validity is the impact on specific pathophysiological pathways known to be associated with the condition. This is currently not readily testable in delirium since the critical neurobiological or pathological pathways have not been defined and could vary between species. Many mechanistic contributors to, and mediators of, delirium have been hypothesized, with perhaps the strongest clinical support for central nervous system inflammatory activation, but overall the specific molecular neurobiology of delirium remains unresolved. Validation of delirium animal models is an ongoing, iterative process that should incorporate as many of these frameworks as possible that become testable over time.

1.3 |. Animal models of delirium: Current state of development

To date, most efforts to model delirium in laboratory animals have used a framework of an underlying vulnerability with a superimposed acute precipitating factor, a framework that has also informed the earliest systematic understanding of human delirium¹⁰ and that has been applied to other multifactorial syndromes of older adults such as frailty or falls.^28,29 Table 3 lists the vulnerabilities and precipitants reported thus far in preclinical models for delirium. Part of the challenge of modeling delirium is the complexity of the clinical syndrome with its multifactorial etiology, both physiological and environmental. As in many manifestations of brain function, both normal and abnormal, delirium can best be considered an emergent behavior of a complex system.³⁰ The heterogeneity among the vulnerability and precipitating factors likely reflects the multifactorial nature of delirium and must be considered when modeling a complex clinical syndrome such as delirium. Models based on a single intervention, such as apolipoprotein E (APOE) gene knockout, mutant APP transgene overexpression, or high fat diet feeding, may not be as relevant because a single intervention is unlikely to prevent or substantially improve adverse outcomes associated with the condition.

TABLE 3.

Overview of select animal models for delirium

Model description (animal)	Reference	Age	Vulnerability factor	Precipitant	Model details/effects
Female C57BL/6 Mice	125	6–7 months	None	Urinary tract infection	Frontal and hippocampal functional impairments, specifically increased thigmotaxis in open field, and reduced spontaneous alternations in Y-maze correlating with neuronal injury marker, cleaved caspase-3
C57BL/6-ME7 Mice	40	8–12 weeks	1 microliter of 10% w/v scrapie (ME7 strain)-infected C57BL6 brain homogenate	100 μg/kg of LPS IP	Extracellular amyloidosis, synaptic loss, robust neurodegeneration, and progressive cognitive decline; diminished performance on T-maze test performed at 12 weeks post-disease initiation
C57BL/6 - hypocholinergic model Mice	23	Not stated	Preexisting neurodegeneration of basal forebrain cholinergic system	100 μg/kg LPS	Depletion of cholinergic neurons in basal forebrain and decreased innervation of hippocampus; diminished performance on T-maze test performed at 1 month post-cholinergic lesioning
Male Wistar Rats	126	Adult	None	IV administration of anticholinergic drug atropine	Impaired attention and memory, sleep-wake cycle reversal, and impaired focus, increased irritability, and fluctuating levels of activity
C57BL/6-ME7 Mice	127	Not stated	One μl of a 10% w/v scrapie (ME7 strain)-infected C57BL/6 brain homogenate	100 μg/kg of LPS IP	Y-maze performed 12 weeks post-disease initiation. Diminished performance on Y-maze test, motor tasks, and suppressed locomotor test; acutely impaired motor coordination and muscle strength, earlier onset of disease-associated impairments
Wistar rats	128	25 months	Age	Clamping upper mesenteric artery under sevoflurane anesthesia	Memory and learning impairment after surgery
Sprague-Dawley rats	129	20months	Age	Anesthesia only 4 h isoflurane 1.3%	Memory and learning impairment
Sprague-Dawley rats	130	20months	Age	Laparotomy under isoflurane anesthesia	Memory and learning impairment, diminished performance on Morris water maze test
Sprague-Dawley rats	131	22–23 months	Age	Splenectomy under isoflurane anesthesia	Memory and learning impairments, diminished performance on Morris water maze reversal test
Sprague-Dawley rats	132	18 months	Age	Laparotomy	Impaired spatial memory and diminished performance on Morris water maze (latency+ distance)
Fischer 344 rats	33	24 months	Age	50 μg/kg of LPS IP	Impaired reversal learning and attentional shifts without affecting discrimination learning in attentional set-shifting task, indicating deficit in attention and cognitive flexibility
Male BALB/cmice	34	20–24 months	Age	0.33 mg/kg of LPS	Neurobehavioral impairments consistent with delirium, depression, and early onset of neurodegeneration
Female C57BL/6J mice	37	18 months	Age	Anesthesia, laparotomy	Cognitive impairment was measured using Morris water maze, and Barnes maze
Female C57BL/6 mice	38	18 months	Age	Laparotomy under isoflurane anesthesia	Time-dependent impairment in buried food test, open field, and learned behaviors (Y-maze test) induced by surgery/anesthesia in aged mice
Female C57BL/6J-ME7 mice	41	5–8 months	One microl of 10% w/v scrapie (ME7 strain)-infected C57BL6 brain homogenate	100 or 250 μg/kg of LPS IP	Impaired locomotor and cognitive function (open field test and escape-from-water paddling T-maze)
MaleC57BL/6 mice	44	12 weeks	Age	Tibial facture under isoflurane anesthesia	Five-choice serial reaction time task as a measure of inattention
Male C57BL/6 and Ccr2RFP/+Cx3cr1GFP/+ mice	45	10–14 weeks	Age	Tibial facture under isoflurane anesthesia	Glial activation, macrophage infiltration, cytokine release, and blood-brain barrier permeability
C57BL/6J mice	46	4 months	Age	Simple laparotomy under isoflurane anesthesia	Buried food test, open field test, and Y-maze test
C57BL/6J mice	47	2–8 months	Age	Simple laparotomy under isoflurane anesthesia	Open field test at 24 h following surgery plus anesthesia and increased a-synuclein and S100β
Female C57BL/6J mice	42	8–12 weeks	One microl of 10% w/v scrapie (ME7 strain)-infected C57BL6 brain homogenate	LPS	Food reward and escape from water motivated T-maze task
Male C57BL/6J mice	35	18–20 months	Age	Sevoflurane anesthesia, laparotomy and 12 h of ICU	Acute impairment of attention, memory, and thought organization; all mice were tested in Y-maze novel arm preference, buried food, attentional, and open field tests
Female mice	36	9–18 months	Age	Anesthesia and abdominal surgery	Delirium-like behavior assessed via buried food, open field, and Y-maze tests
C57BL/6-ME7 mice	39	Not stated	One microl of 10% w/v scrapie (ME7 strain)-infected C57BL6 brain homogenate	LPS, 100 μg/kg	T maze performed 12 weeks and 15 weeks post-disease initiation. Progressively diminished performance on T-maze test (fluctuating signs) due to progressive underlying degenerative features
MaleC57BL/6 mice	48	6–10 weeks	Age	Streptococcus pneumoniae pneumonia-induced sepsis	Behavioral tests include open field, Y-maze, elevated plus maze, Morris water maze, immediate shock deficit, cued and contextual fear conditioning
APPSwe/PS1dE9 mice	133	5–22 months	Double transgenic mice expressing a chimeric mouse/human amyloid precursor protein and a mutant human presenilin 1 (PS1-dE9), both directed to CNS neurons; both mutations are associated with early-onset Alzheimer’s disease	100 μg/kg of LPS IP	LPS challenge acutely disrupted cognitive flexibility/executive function in Y-maze reversal learning tests and suppressed locomotor activity (performed at 15–16 months)
C57BL/6J Mice	134	24 months	Age	Poly I:C administration (12 mg/kg i.p.)	Acute disruption of working memory as measured by the “escape from shallow water” version of T-maze alternation in animals challenged with poly I:C
Per2-luck knock-in mice (C57BL/6J background)	32	4–8 months	Chronic disruption of circadian timing	LPS i.p. injection	Altered clock gene expression rhythms in suprachiasmatic nucleus, thymus, and peritoneal macrophages; no behavioral measures of anxiety/depression after circadian disruption
APPDwDI/mNos2−/− (CVNβAD) mice	43	3 and 12 month	AD-vulnerable brains (CVNβAD mice)	Tibial fracture surgery under isoflurane anesthesia	Failed resolution of inflammation, increased neurovascular vulnerability, hippocampal Aβ deposition, and impaired attention in 5-CSRTT
Sprague-Dawley rats	49	14–16 weeks	None	Cardiopulmonary bypass with deep hypothermic circulatory arrest	Reduced microglia activation, reduced cell death, and improved neurological motor scores after treatment with Annexin A1
Swiss Webster mice	50	3 month	Night time dim light exposure	LPS injection	Disrupted microglia inflammatory response, exaggerated sickness response with decreased home cage locomotor activity (on day following LPS)
C57BL/6J mice	53	99 weeks	Age	Polypharmacy diet	Declines in locomotor activity, rotarod latency, and front paw wire holding impulse

Abbreviations: CNS, central nervous system; LPS, lipopolysaccharide.

1.3.1 |. Vulnerability factors

Because aging and neurodegenerative disease are primary nonmodifiable risk factors for human delirium, they have often been employed as vulnerability factors in rodent models. Wild-type (WT) mice or rats are aged up to 24 months (approximate median lifespan) for aging-related research.^31–38 Neurodegenerative models used in the delirium literature include the ME7 prion model,^39–42 APPSwDI/mNos2^−/− AD model,⁴³ and murine-p75-saporin immunotoxin (mu-p75-sap) to induce selective lesions of the basal forebrain cholinergic system in mice.²³ Many other neurodegeneration models exist, and much work remains to integrate the diversity of available models to better understand the mechanisms that predispose a person to delirium and dementia. It is also important to note that vulnerability factors are not necessarily required in animal models, as delirium has sometimes been observed in their absence, particularly when the precipitant is sufficiently severe (eg, sepsis).

1.3.2 |. Precipitating factors

Superimposed on any baseline vulnerabilities are precipitating factors that attempt to mimic common acute events or stresses that lead to delirium in humans. This can take either a reductionist or emulative approach, depending on the experimental goals. An emulative approach attempts to mimic the pathology of delirium by simulating a relevant multifactorial clinical scenario. Models used include experimental tibia fracture and repair under anesthesia and analgesia;^43–45 anesthesia and laparotomy;^36,38,46,47 mechanical ventilation-induced acute lung injury, Escherichia coli urinary tract infection, intratracheal administration of Streptococcus pneumoniae to cause pneumonia, sepsis, and multiorgan failure;⁴⁸ cardiopulmonary bypass under anesthesia with hypothermic cardiac arrest,⁴⁹ and the combination of anesthesia, laparotomy, and simulated ICU environment.³⁵ The reductionist approach adopts a more specific precipitating factor, usually in the context of mechanistic studies involving that factor. Examples include bacterial lipopolysaccharide (LPS),^{32–34,39–41,50,51} interleukin (IL)-1β,⁴¹ intracerebroventricular Aβ oligomer injection,⁵⁵ and intracerebroventricular albumin infusion.³¹ Caution with interpretation should be exercised in that some of these precipitants, LPS in particular, produce physiological changes that are independently associated with delirium, such as hypoglycemia⁴¹ and localized cerebral hypoxia.⁵²

Many of the elements that have been implicated as contributors to or precipitants of delirium (eg, polypharmacy,^53,54 sleep disruption^32,50,55) and that inform the core of effective multicomponent delirium prevention interventions have not yet been specifically studied in a delirium preclinical context. Other delirium-associated elements ready for investigation include exacerbating factors (eg, frailty, hypoxia, sensory deprivation, dehydration, and social isolation) and protective factors (eg, mobility, exercise, nutrition, cognitive stimulation, and socialization).

1.4 |. Limitations of preclinical models for delirium and next steps for development

1.4.1 |. General principles

First, it is irrational to expect that animals will experience the full complexity or exhibit the characteristics of delirium that are well documented in humans.⁵⁶ Second, like many neuropsychiatric disorders, the constellation of features present in the human syndrome of delirium are not easily quantifiable or disentangled in small animals (eg, inattention, disorientation, hallucinations), though these are individually nondiagnostic. Given these challenges, modeling of delirium in preclinical models requires recreating clinically known risk factors, for example advanced age, inflammation, or neurodegeneration, with clearly defined behavioral endpoints. To the extent that these behaviors reflect those of human delirium, the underlying neurobiology can be explored.

As previously stated, it is unrealistic to expect any animal model to recapitulate all required features of the clinical diagnosis of delirium. Without knowing the underlying neurophysiology, the best we can currently do is develop a set of animal models with individual cognitive and behavioral features resembling human delirium, and then dissect and explore the common biological mechanisms. This distinction between an animal model with individual features of human delirium versus recapitulating the full repertoire of features of human delirium in animals is important, but it does not negate the utility of animal models. We are very limited in our ability to determine the underlying neurobiology of delirium in humans, so animal models are necessary in order to explore mechanisms and to discover and test therapeutic targets and/or interventions for delirium and its associated adverse outcomes.

We acknowledge up front that a major challenge for future work in the development of preclinical models will be needed to determine the best approach to capture delirium in phenotypic and behavioral tests. Many questions remain unanswered at this time: How do we integrate individual tests in a battery? How do we account for key factors such as timing and sequence? And how should the data be aggregated and analyzed? These questions underscore the pressing need to advance methodological approaches to move the field forward.

1.4.2 |. Realistic goals for preclinical models

Rodents, and particularly in-bred laboratory mice, are the most common preclinical model for many diseases and disorders. Examples abound, from depression and schizophrenia to atherosclerotic coronary disease and AD/ADRD. Despite this, few of these syndromic diseases map precisely from human to mouse.⁵⁷ Thus, we recommend a cautious approach that considers the limitations of developing preclinical models for delirium.

Several potential approaches for the creation of preclinical models were considered. The first approach uses knowledge of the genetic basis of human disease to create genetically modified mice. This has included knocking out the low-density lipoprotein (LDL) receptor or apoE lipoprotein to create models of hypercholesterolemia-driven atherosclerosis⁵⁸ or overexpressing APP transgenes with human familial AD mutations to create models of AD.⁵⁹ Such approaches are currently unavailable for delirium, which is unlikely to have a monogenic inherited predisposition.

A second approach, available in cases where mouse physiology is similar to that of humans, involves mimicking the conditions that elicit the human disease. For example, investigators have used high-fat/high-sugar feeding of sedentary mice to mimic many of the features of human obesity, type 2 diabetes, and nonalcoholic fatty liver disease.⁶⁰ In delirium research, this approach is being explored by subjecting mice to surgery with anesthesia, systemic infections, or inflammatory processes, factors that often trigger delirium in humans.

A third approach is based on the idea that the underlying cellular and molecular neurobiology and pathology may indeed be shared by mice. This includes activation of certain immune cells,⁶¹ or the disruption of the blood-brain barrier integrity should be accompanied by behavioral features resembling human delirium.⁶² However, we again emphasize that key clinical features of delirium are both nonspecific and difficult to quantify, so many behavioral features associated with these interventions may be mistaken for those of delirium if factors confounding the tests are not carefully controlled for.

A fourth, and possibly more realistic, goal of the preclinical model is to interrogate specific biological mechanisms rather than recreate delirium in toto. Though this approach has been utilized for AD/ADRD with the use of transgenic APP models, it is important to note that these mice do not have AD/ADRD per se. As such, efforts to make them more directly and translationally relevant have included multigenic models and aged knock-in models. Preclinical models for delirium face a higher bar and greater challenges than for preclinical models of AD/ADRD given an absence of (a) a familial, hereditary-based component and (b) objective, validated neurobiology or pathology of delirium to guide potential mechanistic investigations.

A fifth approach relies on findings that some molecular and cellular responses to inflammatory or metabolic insults seem to be conserved across species,⁶³ suggesting an opportunity for validation in vitro and with empirical evidence from human biomarker studies. The cognitive effects of these insults in rodents does not need to precisely mimic delirium in humans to provide useful information. In other words, mechanisms hypothesized to contribute to delirium in humans may produce phenotypes in rodents that approximate features of human delirium. With this approach, the face validity of these phenotypes is improved if shown that the phenotypes in rodents are consistent with that of human delirium (eg, being transient, affecting attention, and higher prevalence or greater severity in older animals and in animals with underlying neurodegeneration).

Lastly, although current murine delirium models have illustrated moderate utility, improvement within the models requires further cross validation between clinical and preclinical research. For example, leveraging clinical specimens and established biomarkers from well-phenotyped human cohorts can be combined with preclinical models to further clarify the role of specific biomarkers involved in delirium.⁶⁴ In addition, clinical studies may identify genetic predispositions to inform model creation and distinguish among mechanistic subtypes or mechanism-phenotype associations. When revealed, effective pharmacological treatments for delirium in humans or treatments that disrupt the delirium-neurodegeneration link present powerful opportunities for validating models that may ultimately have clinical benefit. Work in animal models more closely related to humans (eg, in nonhuman primates [NHP]), may provide enhanced opportunities to obtain mechanistic data on critical pathways and targets, as opposed to studies conducted in more phylogenetically distant organisms (eg, mice). We acknowledge that capturing complex, human behaviors is more realistic in larger mammals, in particular in NHP; however, logistic or cost constraints may limit such studies.

1.5 |. Phenotypic and behavioral assessments of delirium features in animal models

Human delirium is a clinical diagnosis that relies on a spectrum of observable behaviors and their timing. Many of these behaviors are compared against what the observer considers to be normal behavior. Similarly, a deviation in typical behavior, such as an acute change in “normal” behavior in rodents, for example, may be an indication of delirium, although such deviations in behaviors may not necessarily mimic delirium in humans. As a starting point, one could use the Confusion Assessment Method (CAM) criteria⁶⁵ (widely applied in human delirium), for which specific domains can be tested in rodents. Noting the important DSM-5 criterion that the observed behavior “is not better explained by a preexisting, established, or evolving neurocognitive disorder,” it is important to verify that observed mouse behavioral features are not assigned to delirium or to cognitive impairment if they are better explained by altered locomotor activity, reduced motivation, increased stress, or other aspects of sickness behavior.⁶⁶ Table 4 lists example behavioral tests (further discussed below) for rodents.

TABLE 4.

Examples of rodent behavioral tests relevant to delirium

Human delirium feature	Rodent behavioral test examplesa	References
Acute onset and fluctuating course	Assessment of behavior pre- and postinduction of delirium-inducing condition, home cage monitoring	32,34,50
Inattention	Spontaneous alternation (Y-maze), novel object interest/recognition, acoustic startle response, pre-pulse inhibition of startle response, 5-choice serial reaction time task, attentional set shifting	33,35,43,44,125
Disorganized thinking	Nest building, attentional set shifting task, T-mazea	23,33,35,39,40
Altered level of consciousness	Open field, novel object interest/recognition, acoustic startle response and prepulse inhibition of startle response, home cage monitoring	41,50,55,125
Memory impairment	Spontaneous alternation (Y-maze), T-maze, novel object interest/recognition, nest building, buried food test	35,36,38,40,55,125
Psychomotor activity (agitation or retardation)	Open field to assess time spent in center or periphery, elevated plus maze, home cage monitoring	50,125
Altered sleep-wake cycle	Home cage monitoring	50,55

^aFor several of the human delirium features, no one test currently exists that reliably informs the human delirium feature for which it is currently listed. We include these tests as potential examples for which a battery of tests is required to reveal the possibility of a common defect within the human delirium feature from expected behavior since the presence of a single test does not equate to the presence of a given delirium feature in rodents.

1.5.1 |. Key delirium features

Acute onset and fluctuating course.

Acute onset and fluctuating course is a central criterion for the diagnosis of delirium. This is often difficult to assess in animals. The “acute” refers to the proximity of the behavioral change to a precipitant (eg, surgery or drug) and, to a lesser extent, the suddenness of the change. Frequent repeat testing is difficult because careful behavioral measurements can take hours and learning effects should be considered. Thus, the ease with which we can detect behavioral changes that are “acute” is diminished. However, changes that genuinely represent an acute onset change from baseline measures on the relevant parameters, such as working memory and attention, are possible to detect and have been performed.⁴⁴ Similarly, a “fluctuating course” requires repetitive evaluations with variable outcomes. This requires tracking individual animals rather than analyzing groups of animals (as is often done in animal research). Such approaches to detecting a fluctuating course have been reported.³⁵ While challenging to detect, demonstrating acute and fluctuating course is a key feature of delirium and reflects an important objective for the field.

Inattention.

“Inattentiveness” is another central criterion in the diagnosis of delirium in humans. Some rodent behavioral assays are reported to capture attention. This includes the five-choice serial reaction time task (5-CSRTT),⁴³ attentional set shifting,⁴⁰ and buried food test.^36,37 More recent work has stressed the importance of including distracting stimuli so that the animal must remain vigilant for the correct target stimuli. These “continuous performance tasks” have been reverse translated to humans and been shown to capture attentional deficit disorder⁶⁸ and therefore should be useful to assess attention or vigilance in animals. Although some of these tasks might be more specific for attention, in reality, the majority of rodent behavioral assays require a degree of attentiveness, as well as executive function, memory, and sensory abilities (eg, sight, sound, and smell). Even in nonmotivational tests (eg, novel object recognition and open field), the rodent must be attentive to its environment to display behavior that can be scored. Although attention is the only minimum required attribute, the rodent must also recognize the environment as novel or dangerous, which implies that memory and cognition are additional key attributes important for the detection of inattention. Similarly, tests that involve a reward (usually food) probe memory, sensory, and reward pathways that are incrementally more complex behaviors than isolated attention.³⁴ Thus, although a single animal assay for attention remains elusive and is unlikely to appropriately capture inattention, several overlapping tests that require attention may reveal a common defect (ie, change in attention). For example, NHP are capable of several attentional tasks that mirror attention in humans (eg, visual search tasks). In such animal models, it is critical to rule out motivational, appetitive, locomotor, memory consolidation deficit, and severe sickness behavior before concluding that a given test demonstrates inattention.

Disorganized thinking.

In humans, disorganization of thought is determined on the basis of a set of established human norms to establish how an organized train of thought should progress. Although disorganized thinking in an animal might be inferred from its behavior (eg, not seeking food, exploring a familiar rather than a novel object, not being fearful in a known dangerous environment), violation of normative behaviors (eg, grooming a lower-ranking member or stealing food from a higher-ranking member of the troop, as can be observed in NHP) is difficult to objectively measure. This is further compounded by the knowledge that these measures are subject to important confounding factors. As with detecting the presence of inattention (described earlier), it is important that potential confounding factors are assessed to avoid mistaking changes in motivation, anxiety, or memory for disorganized thinking. Some questions to consider include the following: Is the animal motivated to seek or consume food? Is the animal’s lack of fear explained by the failure to consolidate the memory of the fearful event?

Apart from NHP work, the field could additionally benefit from studies involving rodent models subject to assessments of cognitive disorganization (eg, prepulse inhibition, as used in schizophrenia⁶⁹ and traumatic brain injury⁷⁰ models). Though overt psychosis is more challenging to assess in rodents, newer computational-behavioral techniques may be useful for quantifying hallucination-like perception in mice.⁷²

Altered level of consciousness.

Since assessing level of consciousness relies on generally primitive responses to overt stimuli, this is among the least difficult features to assess in an animal model. The speed and completeness of the righting reflex, for example, might be a gauge of intermediate levels of consciousness that could be consistent with, but nonspecific for, delirium. By evaluating an animal’s response to overt stimuli, such as being placed in an unfamiliar environment (eg, open field) or having to interface with an unfamiliar object (eg, novel object recognition), it may be possible to gauge alterations in the level of consciousness. In a similar way, alterations in the level of consciousness may be inferred by changes in established animal behavior norms, such as typical behaviors that can be observed with home cage monitoring (ie, spontaneous locomotor activity, grooming, socializing, and rearing).

Arousal and delirium.

Because fluctuations in the level of arousal above and below one’s baseline are a prominent feature of human delirium, alterations in the brain’s arousal systems have long been implicated in delirium pathogenesis. For example, functional magnetic resonance imaging (fMRI) studies have suggested disruption of the ascending reticular activating system in patients with delirium admitted to a medicine ward.^74,75 Increased noradrenaline release from the locus coeruleus has also been associated with heightened noradrenergic transmission in the prefrontal cortex and hippocampus during delirium hyperactive states and linked to inattention and cognitive dysfunction.^76–79 Additionally, since sleep deprivation powerfully alters arousal level and sleep disturbances are common in hospitalized delirious patients, it will be critical to investigate how triggers such as perioperative sleep dysregulation, aging, and anesthetic/sedative drugs may destabilize wake-sleep transitions and contribute to disordered arousal. To date, many of these questions remain open. Substantial research efforts will be needed to resolve whether a direct causative role of sleep disturbances in delirium exists and the underlying wake-sleep circuitry dysfunction.

Emergence from anesthesia might be utilized as a model of delirium. While emergence in the healthy brain typically requires only minutes, prolonged emergence, common in older adults, may resemble delirium. For example, stage 2 general anesthesia is generally an unconscious, hyperactive state associated with light anesthesia that phenotypically resembles hyperactive delirium.⁸⁰ In animals, it would be possible to maintain low concentrations of anesthetics to prolong such states in order to study them for delirium-like behaviors and biomarkers. Further support for the use of anesthesia in delirium research is that, at the network level, both human and rodent brains undergo pronounced fluctuations in activity at a steady concentration of anesthetic,^81–83 which may closely mimic the fluctuating nature of delirium.

1.5.2 |. Minor features

Memory impairment.

There are many tests for memory, both spatial and contextual, in the rodent. Although the tests are mostly limited to spatial and olfactory memory in rodents, impairments in memory would reasonably be expected to accompany delirium in an animal, though not be specific to it.

Psychomotor agitation or retardation.

Enhanced or diminished motor activity over that normally seen for the time of day or age and sex of the animal might reflect psychomotor changes relevant to delirium and may provide an assay distinguishing hyper- from hypoactive delirium. Sophisticated “smart cages,” in addition to more conventional video-tracking systems, are readily available to monitor spontaneous locomotor activity. In addition, these monitoring systems, which have been adopted for complex analysis of mouse behavior video recordings to examine the impact of therapeutics,⁷³ can measure more elaborate behaviors such as grooming and rearing to assess some aspects of self-care and anxiety, which might also be altered in delirious states. Reduced time spent in the center of an open field may also indicate thigmotaxis, considered a marker of psychomotor agitation. In contrast, avoidance of open arms of the elevated plus maze may suggest the presence of associated anxiety-like behavior (Table 4). It remains important to determine whether motivational, energetic, or stress responses are causal with respect to altered locomotor activity.

Altered sleep-wake cycle.

As mentioned earlier, automated activity monitors have been validated for tracking sleep-wake cycles in rodents. Dysregulated sleep might reasonably be expected to accompany confusional and agitated states and are very common features in delirious patients. The primary advantages here are that (1) sleep is not a learned behavior or a motivational behavior and (2) sleep does not require the physical presence of the investigator. Taken together, these advantages suggest that altered sleep-wake cycles should be free of the usual confounding factors present in many behavioral assays.

Potential pitfalls.

Beyond the obvious challenges of creating a human condition in preclinical models, many nuanced factors must be taken into account when developing models for the study of delirium. The need to assess cognition acutely, paired with the intended and unintended consequences of commonly employed vulnerability and precipitating factors, must be recognized. Because aging and neurodegeneration models are often used as vulnerability conditions in the field, methodical evaluation of these animals should be performed prior to the use of even common behavioral paradigms. For example, such animals are known to develop sensory loss^72,76 and impaired mobility,⁷⁷ which may limit the utility of behavioral paradigms requiring consistent responsiveness to sensory cues (eg, prepulse inhibition and many operant conditioning paradigms) or intact mobility (eg, Morris Water Maze), respectively. The often pleiotropic effects of precipitating factors must be similarly appraised. For example, when utilizing LPS or live-bacterial models of sepsis, loss of appetite and food motivation caused by sickness behavior should be considered. This potential experimental pitfall has been directly addressed through the development of a novel “escape from shallow water” T-maze for studying delirium, which strategically employs shallow water as a non-life-threatening, aversive motivator in an otherwise classic T-wave alternation task testing working memory.⁴⁰

Necessity of behavioral assessment battery.

The implicit assumption of mapping these human phenotypic criteria of delirium to an animal is that they represent a similar neurobiological state. The validity of such an assumption is unknown. The behavioral response to, for example, a significant neuroinflammatory stress might simply be hypoactivity or sickness behavior in both human and animal. But the additional neurobiological vulnerabilities that convert sickness behavior into delirium could be different in animals, or at least have different thresholds. A related consideration is the potential need for normalization based on the animal’s level of activity, which could vary depending on the disease model and confound the assessment of certain behavioral assays.

It is clear that no single animal behavioral assay directly assesses a state analogous to human delirium and, because the cause of particular animal behaviors can only be inferred, that the behaviors may not be specific to a delirium-like state. Many tests, however, probe components of delirium. Attention and memory, as examples, are required in most animal behavior assays. Spontaneous, natural behaviors, such as that in open fields, novel object recognition, and sleep quality, might be included to broaden the behavioral repertoire, and these changes may also offer some explanatory aids to the interpretation of other tasks performed. In addition, if appropriately controlled, motivational assays could get closer to attention and disorganized thinking. Thus, we recommend that investigators who aim to characterize a delirium-like outcome use a battery of tests, when possible, to assess different cognitive, sensory, reflex, arousal, and reward networks. We recognize, however, that many questions regarding the use of these tests remain: How do we integrate individual tests in a battery? How do we account for key factors such as timing and sequence? And how should the data be aggregated and analyzed? These questions underscore a growing need for more advanced methodological approaches to move the field forward.

Finally, the utility of measuring several parameters in each experiment needs to be balanced against the advantage of performing repeat evaluations, at intervals as closely spaced as practical, in order to track acute onset and fluctuating courses. Consistent with the latter, statistical analysis should track individuals, rather than just groups, over time in order to assess whether any findings are acute and fluctuating at an individual level. Imperfect as these tests are, the testing of potential treatments would provide a pathway to the eventual refinement of the models.

Ultimately, it is likely that the identification of the appropriate behavioral battery can only be established by first replicating conditions that have been strongly associated with human delirium (eg, age, inflammation, and drugs) and then characterizing the behavioral outcome, rather than trying to map CAM onto an animal, in other words, prioritizing construct validity rather than face validity. Evaluation of altered consciousness and disorganized thinking requires expert clinical judgment in humans and can be difficult to assess in animals since they are nonverbal. Therefore, an alternative starting point is to take a behavioral neuroscience approach and use well-characterized tasks, which unambiguously interrogate specific cognitive domains, in order to determine which traits are impaired in models with good construct validity. This iterative process should ultimately be useful in identifying behavioral patterns most representative of human delirium. Because no specific drug has been shown to mitigate delirium, using pharmacology to validate the state, as has been successfully done in schizophrenia and depression, seems implausible at the current time. Because delirium animal models continue to evolve, the testing of potentially effective treatments may provide a pathway to the eventual refinement and validation of the models.

1.6 |. Biomarkers of delirium

In human studies to date, many biomarkers temporally associated with delirium have been identified (Table 5), with varying degrees of replication or validation. These biomarkers are presented here since they may provide targets against which animal models may be validated. Iterative refinement in translational bedside-to-bench and bench-to-bedside studies will allow cross-validation of these biomarkers and allow hypothesis generation for delirium neurobiology or pathology. Thus, although the biomarkers in Table 5 must be referred to only as “potential,” they hold promise for cross-validation studies to move the field of delirium forward.

TABLE 5.

Candidate delirium biomarkers for animal model validation

	Humans		Animals		Human evidence for:
Marker	Human evidence	References	Evidence of causal effects in animal models	Refs	Risk markers	Disease markers	Outcome markers
Biofluid Based
Inflammatory
• AGP	↑ CSF	145
• CHI3L1/YKL-40	↑ Blood	146			X	X
• Complement factor C3	↑ Blood, CSF	147
• CCL2	↑ Blood	148				X
• CRP	↑ Blood, CSF	149–151			X	X
• FGF-21, FGF-23	↑ Blood	152				X
• Gal-3BP	↑ Blood	153
• IGF-1	↓ Blood	148,154				X
• IL-1β	↑ Blood, CSF	155	Causal in multiple models			X
• IL-1RA	↑ Blood, CSF	156,160	Protective in some models
• IL-2	↑ Blood	157				X
• IL-6	↑ Blood, CSF	156–162			X	X
• sIL-6R	↑ Blood, CSF	163
• IL-8	↑ Blood, CSF	162				X
• IL-12	↑ CSF	164
• MCP-1	↑ Blood	165				X
• Natural killer (NK) cells	↑ Blood	166
• Syndecan-1	↑ Blood	167				X
• TNF-α	↑ Blood, CSF	168	Causal in some models		X	X
Neurodegeneration or neuronal injury
• Aβ	↑ CSF	169,170
• GFAP	↑ Blood, CSF	170–173				X
• Neurogranin	↑ Blood	174				X
• Neutrophil:Lymphocyte ratio	↑ Blood	175,176			X	X
• NSE	↑ Blood	173,177,178				X
• NfL	↑ Blood	170,171,179				X
• P-tau	↑ CSF	180				X
• PAI-1	↑ Blood	161
• S100β	↑ Blood, CSF	162,178,181				X
• T-tau	↑ Blood, CSF	170,180,182				X
• UCHL1	↓ Blood	183				X
Biochemical:
• AChE	↓ Blood, CSF	168,183,184			X	X
• Adrenaline	↓ CSF	176				X
• Anticholinergic activity	↑ Blood, CSF	185	Disruption of relevantfunctions by cholinergic antagonists
• BuChE	↓ Blood, CSF	186,187				X
• Dopamine	↓ CSF	187				X
• FA5	↑ CSF	188
• Troponin	↑ Blood	175				X
Neurophysiological
• EEG—oscillatory dynamics	Relative power increase in slow frequencies and decrease in fast frequencies	87,89–102,189–192	↑ slow wave activity	107		X
	—↑ θ
	—↑ δ
	—↓ β
	—↓ α
	And loss of high-frequency activity, especially in posterior brain regions					X
	Altered α/θ, α/(θ+δ) or (α+β)/(θ+δ) ratio	94,101,192				X
	Reduced peak frequency	101			X
	High preoperative alpha power	95			X	X
• EEG—connectivity	↓ α band connectivity (possible marker for disturbed consciousness) and loss of directionality	94,102,103,191,192	↓ α band connectivity	107		X
	↑ θ band connectivity (possible marker for attention/memory deficits)	94,192				X
	↓ β band connectivity (possible marker for instability of cognitive processes)	192				x
	↑ δ band connectivity directed to frontal regions	191,192				X
	Disturbance in default mode (DMN), salience (SN), and executive control (ECN) networks	192				X
	Decrease in frontal-occipitoparietal connectivity	95				X
• SSEP/BAEP	↑ Short- and long-latency sensory evoked potentials (SEP)	113				X
Local field potentials (LFPs), Single neuron activity	↓ Hippocampal θ frequency; ↓θ frequency and amplitude in medial prefrontal cortex; ↑ short-term δ frequency	118
Imaging
• Structural (MRI, DTI-MRI)
	↑ Ventricular size (as a vulnerability marker)	193				X
	New ischemic lesions, ↑ white matter hyperdensity/disruption	194–198			X	X	X
	↓ Gray matter volume	196			X
	↓ Brain volume	199					X
• Functional (fMRI)
	↓ Connectivity strength (in aging and dementia)	103			X
	↓ Efficiency and local clustering	103				X
	Imbalanced connectivity in default mode network (DMN) and disrupted connectivity between DMN and dorsolateral prefrontal cortex (executive network)	74,200				X
	Excessive striatum-salience network (anterior cingulate cortex and insula) connectivity	200				X
	Disconnection of mesencephalic tegmentum, ventral tegmental area, and nucleus basalis with striatum/thalamus	74,200	LPS-induced T-maze deficits reversed by peripheral glucose			X	X
• Functional (PET, SPECT)
	↓ Whole-brain glucose consumption, especially in posterior cingulate cortex	200		59
	↓ Blood flow ratios in frontal and parietal cortex	201	↓ glucose consumption in neocortex			X

Abbreviations: AChE, acetylcholinesterase; AGP, alpha-1-Acid glycoprotein; BuChE, butyryl cholinesterase; CRP, C-reactive protein; CSF, cerebrospinal fluid; FA5, coagulation factor V; GFAP, glial fibrillary acidic protein; IGF, insulin-like growth factor; IL, interleukin; CHI3L1/YKL-40, chitinase 3-like protein 1; fMRI, functional magenetic resonance imaging; Gal-3BP, galectin-3 binding protein; LPS, lipopolysaccharide; NfL, neurofilament light chain; MCP-1, monocyte chemoattractant protein-1; NSE, neuron-specific enolase; PAI-1, plasminogen activating inhibitor-1; R, receptor; PET, positron emission tomography; RA, receptor agonist; s, soluble; S100β, S100 calcium binding B; UCHL1, ubiquitin carboxyl-terminal-esterase-L1.

In our usage, “biomarker” is used broadly and includes not only biofluid analytes but also neurophysiological and neuroimaging modalities. Currently, several clinical studies have described a number of biomarkers associated with human delirium; however, their specificity for delirium is not well established. Additionally, whether the association between a biomarker and clinical phenotype represents causality remains unclear. Biochemical, imaging, and neurophysiological markers associated with human delirium and measured in animals with delirium-like behavioral changes will enhance both the face and construct validity of the animal models and will facilitate moving beyond understanding the associations between a given biomarker and delirium to identifying causal mechanisms associated with delirium. Nevertheless, once animal models of delirium-like behavior have been validated, they could be effectively used to identify new biomarkers in animals and subsequently to probe their validity in delirious patients, with the ultimate goal of illuminating the neurobiology and pathology of human delirium.

Delirium biomarkers may be grouped into two broad categories: (1) risk or predisposing markers (ie, present before the onset of delirium, indicating substrates conferring vulnerability) and (2) disease markers (ie, present during delirium, which may point to the mechanisms that directly precipitate or sustain delirium). Additionally, a third family of biomarkers may also be considered: outcome markers that are present after delirium resolves and indicate longer-term injury. It is important to note that some biomarkers may simultaneously belong to more than one category. We emphasize here that there is currently no validated biomarker for delirium. In fact, it is likely that a composite of biomarkers will be needed to provide high predictability.

Emerging evidence suggests that biomarkers that could bring mechanistic insights into delirium include biofluid assays for inflammation and brain cellular stress/injury, neurophysiological, and neuroimaging markers (Table 5). Research is urgently needed to clarify (1) the underlying brain injury mechanisms that trigger or sustain delirium, (2) the dysfunctional neural circuitries that underlie the delirium phenotype, and (3) interactions of biomarkers with clinical risk factors to cause long-term injury. Although it is possible to directly sample the animal brain, this only provides a view of the pathophysiology at a single point in time and cannot be directly cross-validated to the human brain. An important advantage of the biomarkers that will be discussed in what follows is that they can be obtained repetitively, providing a longitudinal view of the pathophysiology in an individual animal and compared directly to humans.

1.6.1 |. Biofluid biomarkers

Biofluid biomarkers can be divided into markers reflecting (1) vulnerability for delirium and (2) the consequences of the precipitant. In the category of vulnerability for delirium markers, this would include markers of neurodegeneration such as those associated with AD/ADRD and perhaps inflammatory mediators that are associated with a variety of smoldering neuroinflammatory disorders. These biomarkers are typically not found in WT animal models, but rodents can be genetically modified and these biomarkers subsequently measured in tissue and biofluids. The category of markers reflecting the consequences of the precipitant can be further divided into a series of inflammatory mediators (reviewed in Reference 19), such as interleukin-6 (IL-6), tumor necrosis factor-alpha (TNFα), and high mobility group box protein-1 (HMGB1), markers readily measured in animals (see Table 5 for additional details and citations). These mediators may also cause neuronal injury, reflected by more durable biomarkers, similar to those predisposing to delirium vulnerability. In fact, this might be why an episode of delirium increases the risk of future delirium episodes.

Validating whether these biomarkers are causally linked to delirium remains a challenge since in humans such linkages are associative at best. Although controversial, it is plausible that delirium may reflect neuronal injury rather than a transient imbalance in neurotransmitter release or bioenergetics across networks and brain regions. Many neuronal injury biomarkers exist, including glial fibrillary acid protein (GFAP), tau (and its isoforms), neurofilament light chain (NfL), and glial-derived protein S100 beta (S100β). Of note, however, some neuronal injury biomarkers, such as NfL and tau, are also found in abundance peripherally. Thus, if the delirium precipitant is direct tissue injury (surgery, trauma) or results in indirect tissue injury via hypoperfusion (hypovolemia, sepsis), then these and other biomarkers may reflect the severity of peripheral injury and not necessarily injury to the brain. Therefore, the attempted association of these biomarkers with delirium would have only indirect validity. This has been demonstrated in humans, where plasma NfL but not cerebrospinal fluid (CSF) NfL increased significantly after orthopedic surgery.⁸⁴ Likewise, proinflammatory mediators will be produced in the blood in all individuals after infection, injury, or surgery, indicating that these markers cannot be considered reliable biomarkers for delirium. These issues are mitigated in animals since blood, CSF, and brain tissue can each be available for assay, and once association is established, causation can be probed with appropriate interventions.

New generations of biofluid biomarkers might be of value in delirium. For example, microRNAs (miRNAs) are small, noncoding RNAs that act as rapid-response regulatory factors and have been investigated as brain injury biomarkers in a variety of settings. In a study of rugby players, a panel of plasma miRNAs linked to inflammation and to vascular integrity predicted with 95% accuracy which players with head injury would develop a concussion.⁸⁵ Though the miRNA source cannot be confirmed, the association is intriguing and may be translatable to animals.

Another novel biofluid biomarker includes plasma exosomes. Exosome “cargo” can contain proteins, mRNA, miRNA, and transcription factors, which provide a snapshot of activity in the cells and brain regions extruding the exosomes. For this reason, exosomes are under investigation as biomarkers of, for example, AD.⁸⁷ Comparison of brain-derived exosomal cargos or miRNAs in the blood of patients with and without delirium might provide novel delirium biomarkers, as well as potential targets for study in animals.

1.6.2 |. Neurophysiological markers

Electroencephalography (EEG)

Accumulating human data support electrophysiological monitoring as a promising biomarker of delirium.^3,87,88 Electroencephalography (EEG) can be used as an onsite diagnostic tool, especially useful for hypoactive delirium or when communication is compromised in patients recovering from general anesthesia or sedated in the ICU. Given the high temporal resolution, continuous EEG could help define the acute onset and fluctuating course of delirium, one of the key clinical features in the CAM.⁶⁵ It is important to note that (1) the neurophysiological changes accompanying delirium could represent a common neurobiological consequence of multiple divergent pathologies of delirium,^87,89 as well as other neurophysiologic states, and (2) the exact relationship between EEG/sleep-wake disturbances and delirium remains unclear (ie, whether sleep-wake disturbances are a consequence of or a predisposing factor to delirium).

Because of the largely human nature of the delirium syndrome, it will be imperative to cross-validate EEG studies in animals with those in humans (and vice versa), when feasible. In this iterative framework, animal EEG studies that mirror EEG metrics of delirium in humans will be helpful in the validation of those animal models. In turn, EEG studies in reliable animal models for delirium may help identify novel biomarkers that relate to specific mechanistic pathways and, thus, may be useful in extending human clinical findings. With this in mind, the following sections describe electrophysiological and imaging studies of human delirium that in our view can be conducted in animals and would be beneficial, though to-date have received little attention.

Diffuse slowing of the EEG has been described in delirious patients for decades^89–92 and has several advantages. Delirium is associated with an increase in delta (<4 Hz) and theta (4 to 8 Hz) osccillations and a decrease in alpha (>8 Hz) oscillations on the EEG.^93–100 These changes correlate with cognitive performance and delirium severity.¹⁰¹ In fact, one study reported that quantitative EEG could distinguish delirious from nondelirious patients with a sensitivity of 100% and a specificity of 99%.⁹⁶ An additional advantage of EEG is that it provides measures of functional connectivity, which has also been shown to be impaired and less integrated in delirious patients following stressors of various types.^95,102,103 Lastly, a final advantage relates to the relative ease of measuring electrical activity in awake and unrestrained animals. Taken together, these advantages provide a strong rationale for the use of EEG in animal model validation.

Neurophysiological signatures similar to delirium also occur at loss and return of consciousness during anesthesia and recovery, and these could be explored in animals to shed light on the commonly observed altered level of consciousness in delirium.^81,104–106 However, caution is needed because different anesthetics induce distinct neurophysiological changes^81,104–106 and, thus, may not precisely mimic the neurophysiological correlates of altered levels of consciousness during delirium. In addition, these neurophysiological markers may differ depending on the etiology of delirium.^87,89 Nevertheless, the animal models and experimental paradigm that have been used to investigate anesthesia-induced altered levels of consciousness might prove to be of value if adapted to studies of delirium.

EEG features observed in delirium require further validation for (1) temporal correlation with behavioral features, (2) association with the presumed pathophysiological mechanisms of delirium, and (3) distinction from or agreement with EEG features of other neurocognitive disorders that share some behavioral aspects with delirium. Recently, LPS-induced inflammation was found to produce slow waves and decrease anteroposterior connectivity in the electroencephalogram, along with behavioral quiescence and elevated cortical cytokines in aged mice.¹⁰⁷ These data could then be considered consistent with hypoactive delirium and support the use of EEG markers to distinguish delirium types in animals. To establish EEG markers for all types of delirium including hyperactive and mixed type, it will be essential to demonstrate temporal correlations between these markers and delirium behaviors in the model. Although these behaviors can be challenging to interpret in rodents (see preceding discussion), human-like hallucinatory and locomotor behaviors are often present in NHP models.^108–110 Further, reliable cognitive task performance that enables testing of attention, altered levels of consciousness, and dis-orientation is possible in NHP,^104–106 and NHP have demonstrated that recordings of EEG across the neocortex is comparable to those seen in humans.^111,112

Evoked potentials

Additionally, evoked potentials may be used as a neurophysiological marker for delirium. A recent prospective cohort study applied somatosensory (SSEP) and brainstem auditory (BAEP) evoked potentials in deeply sedated critically ill patients.¹¹³ The results suggest that early impairment of the intrapontine conduction time is associated with postsedation delirium, pointing to the possible impairment of the ascending reticular activation system in the pathogenesis of delirium. SSEP and BAEP can be a useful complementary tool for delirium in high-risk patients, but they have yet to be tested in animal models.

Invasive electrophysiology

Direct, more invasive neurophysiological activities, such as local field potentials, single neuron activity, and brain microdialysis (directly measuring neurotransmitter levels), could provide critical insights into the neurophysiological mechanisms of delirium. While becoming possible in human cohorts (eg, electrocorticography recordings prior to epilepsy resection and brain-computer interface studies^114–118), direct neural recordings are primarily applicable only to preclinical animal models. In particular, the brain regions that represent specific cognitive functions or neurotransmitter systems can be directly tested in animal models.^87,89 Note that, to date, direct neurophysiological recordings from animal models of delirium remain very limited.¹¹⁹

1.6.3 |. Imaging markers

Imaging modalities offer two main advantages in the study of delirium pathophysiology in animal models. First, imaging can serve as a bridge between delirium in human subjects, NHP, and smaller animal species. Behavioral data and imaging markers from humans and NHP could guide studies in smaller animals (eg, rodents with mechanistic, tissue-level confirmation that cannot be performed in humans or NHP). Another important advantage of imaging is that certain brain regions that have been associated with delirium cannot be readily assessed through specific behavioral, electrophysiological, or biochemical tests but can be evaluated via imaging. For example, the posterior cingulate cortex, a component of the default mode network that has been implicated in human delirium, is readily visualized with positron emission tomography (PET).

Functional magnetic resonance imaging

Human fMRI studies suggest disruption in the central executive network (active at external task demands) and the default mode network (active at wakeful rest) in patients with delirium. Both networks involve the prefrontal cortex^74,114 and are consistent with the network dysconnectivity hypothesis of delirium.^87,89 These region-specific hypotheses for delirium could be tested in animal models with direct neurophysiological recording; however, unlike EEG, the low temporal resolution of most types of brain imaging require that the animal be motionless for long periods, implying the use of anesthetics, which could confound the results.

Positron emission tomography

A number of imaging modalities exist with the potential to shed light on both anatomic and circuit changes in delirium. However, practical considerations have limited the use of many of these modalities and, thus, have constrained validation of associated imaging biomarkers. PET uses a shorter imaging paradigm, and most of its tracers are minimally affected by a brief sedation since some interrogate processes that are not being modulated rapidly (eg, microglial activation). Another compelling aspect of PET imaging is the ability to identify and create novel probes for the process, receptor, or other desired target. A number of PET tracers, either for clinical research use or for clinical use in humans, are in the late stages of development or nearing approval and are able to detect neuroinflammation, white matter tract injury, apoptosis, autophagy, and changes in metabolism in human subjects with delirium. These tracers include sphingosine-1-phosphate receptor 1, purinoceptor 7, and peripheral-type benzodiazepine receptor 28 for microglial activation, alongside reactive tracers that indicate innate immune activation in the brain (M1, ROS-Trace). These tracers may be able to provide powerful information regarding inflammation, stress, and neuronal/microglia injury processes that supports or complements delirium-like behavioral changes and other neurophysiological markers of delirium in the same animal model.

In summary, the aforementioned biomarkers are highlighted as ones having significant potential to allow cross-validation of neurobiology and pathology between humans and animals. This would not only enable greater face validity of behavioral assays but would also provide a window into the underlying mechanisms of delirium.

1.7 |. Recommendations for preclinical and translational models for delirium

1.7.1 |. General

General guidelines for animal experimental design were recently published (Planning Research and Experimental Procedures on Animals: Recommendations for Excellence [PREPARE])¹¹⁹ and are relevant to preclinical delirium research. As of this writing, an animal model that recapitulates human delirium with high fidelity has not been reported, so it is important for the investigator to clearly specify what feature of human delirium is being studied and the underlying premise. How is this feature in the rodent (or other animal model) similar to that in human delirium, and is the underlying physiology, where known, similar? In what ways is it different, and does this challenge the relevance to delirium? While it will remain unclear whether such a reductionistic approach (ie, feature by feature) is valid for a complex syndrome like delirium, it nevertheless recognizes the imperfect translation of human delirium to rodents (among other animal models). The following guidelines and suggestions are not intended to be proscriptive but rather represent what we consider to be best practices to cautiously advance this type of preclinical research. Given that there is much room for growth in the field, our guidelines and recommendations reflect what is lacking in the literature, and we anticipate that future, more specific guidelines, building on PREPARE,¹¹⁹ will be adopted once the field of animal models for delirium has undergone further advances.

1.7.2 |. Investigator expertise

Animal behavior can be as varied as that of humans and be surprisingly responsive to environmental cues. Thus, it is imperative that investigators be exceptionally consistent in their handling of animals and in their administration of behavioral tasks. This can extend, for example, from the time of day to the identity and gender of the investigator, and even to the clothes they wear.^120,121 Thus, many investigators are now turning to animal behavior core labs to administer tests and collect data in order to enhance reproducibility.

1.7.3 |. Species, strain, and sex

The choice of animal species always reflects a balance between cost, regulation, and relevance. Rodents (specifically mice) probably represent the optimal balance, but even here, it can be expensive to maintain colonies to an appropriately vulnerable age. Some mouse behavioral paradigms are validated and standardized, and the relatively low cost allows relatively large sample sizes. Furthermore, when pathophysiological hypotheses for delirium can be formulated, the relative ease of validation based on genetic manipulation is a significant strength of the murine model. Delirium occurs in both male and female humans, so both sexes should be studied in animal studies.

Mice are available in a bewildering variety of largely inbred strains (eg, C57BL/6). Inbred strains offer the advantage of genetic similarity, enabling the elimination of genetic diversity as a confounder in experimental studies. Genetic drift reduces this advantage after about 20 generations, so these inbred strains require “refreshing.” Genetic drift also occurs in transgenic or knock-out (knock-in) animals, potentially reducing or eliminating the hoped-for vulnerability factors. Although genetic similarity is an advantage to the experimentalist in one sense, it is a disadvantage when considering that the human we are trying to model is enormously diverse. It should be recalled that not all older adults enter a state of delirium, likely reflecting genetic and other sources of diversity. Outbred mice, such as the diversity outbred (DO) mouse,¹²² are preferred in this case, but the investigator should be prepared for greater variability in the results, which suggests that greater numbers of animals are needed. Finally, in general, mice are exceptionally resilient to all manner of insults, though it could be expected that outbred strains and WT animals would be more so.

1.7.4 |. Vulnerability

Young, healthy humans, like young, healthy WT animals, rarely enter a state of delirium (ie, are not considered vulnerable). We therefore consider it essential to introduce a vulnerability factor into the model to enhance construct validity. As summarized earlier and in Table 6, this can be advanced age (we suggest 18–24 months), genetic changes (eg, introduction of human AD/ADRD-associated genes), drugs or toxins, polypharmacy, or some combination. We also advocate that the experimental design include control groups of animals without the vulnerability factor to demonstrate its independent role, especially when evaluating underlying pathophysiology.

TABLE 6.

Guidelines and recommendations for preclinical and translational models for delirium

Development of delirium model	Choose species, include both sexes. Choose delirium feature(s) to study, recognizing that high-fidelity models that mimic human delirium are not yet available. Consider face and construct validity in development of model (ie, appropriate age, vulnerability, and precipitating factors [Table 2]) Consider behavioral test(s) and frequency to mimic components of delirium in humans (Table 3). Multiple animal models with different precipitants across different settings maybe required for cross-validation. Carefully control for confounding factors (appetitive, motivational, locomotor, stress-related, illness-induced) that could be inappropriately conflated with delirium-like effects Adhere to current standards for ethical treatment of animals
Experimental framework	Appropriate blinding and randomization of outcome assessments Reproducibility, which requires standardization of protocols and adequate sample sizes Monitoringand physiological control of perturbations, especially anesthesia and infection Standardization of behavioral protocols Consideration of temporal alignment of pathophysiology and behavior
Validation of animal model	Cross-validation of behavioral changes with both human and animal biomarkers (eg, biofluid, EEG, imaging) Testing of animal responses to efficacious pharmacological and nonpharmacological human interventions
Reporting of animal studies	Adherence to PREPARE guidelines119 Standardization and protocol validation Data deposition in publicly accessible databases
Future research priorities	Iterative testing and cross-translation of animal models and human delirium to optimize models, clarify pathophysiological mechanisms, and develop and test therapeutics Large-scale collaborative studies to develop, standardize, and validate animal models for delirium; subsequent alignment with human studies as data become available

Abbreviation: EEG, electroencephalogram; PREPARE, Planning Research and Experimental Procedures on Animals: Recommendations for Excellence

1.7.5 |. Precipitant

Because of the acute nature of most human delirium, it is widely assumed that some precipitant superimposed on the given vulnerability is required to trigger a state of delirium. Reproducing the multiple factors that precipitate delirium in a clinical setting complicates the study design (and feasibility), especially when working with older and frail animals. Nevertheless, one or more such factors could be objectively introduced into the experimental design (eg, surgery and/or anesthesia, infection, injection of inflammatory mediators, and environmental alteration, such as noise, movement, lights, socialization, sleep disruption, drugs, or pain). The resulting behavioral phenotype should be monitored closely and repeatedly at intervals designed to capture the “fluctuating” nature of the behavioral change. Some precipitants may be so stressful or disruptive (eg, sepsis) that a state of delirium could be entered without the need for a vulnerability. Also, since some precipitants may be physiologically disruptive (eg, anesthesia or LPS injection), we consider it important to monitor and report physiology (eg, temperature, heart and ventilation rate, oxygen saturation, and blood pressure) to the extent possible in the animal. While it remains contentious in humans, some animal studies have indicated that some general anesthetics alone can cause subsequent acute cognitive disturbances, as well as neurodegenerative features.^37,87 However, it is critical to note that general anesthesia in a rodent is not analogous to that in a human patient. Patients undergoing anesthesia have their physiology carefully and continually monitored while rodents are only superficially monitored, if at all. In fact, in some rodent studies, prolonged, multidrug, repeated anesthetic exposures can cause sufficient physiological disturbance (eg, hypoxemia, hypotension, or hypothermia) to independently cause cognitive disturbances.

1.7.6 |. Outcome measures and validation

Two types of outcome measures should be considered in the experimental design. First, the behavioral phenotype is crucial, as it is (currently) the primary means of alignment with the human condition. As discussed earlier, human delirium involves several behavioral domains (from awareness to motor), but what is measured does not necessarily need to encompass every domain, depending on the goals of the experiment. In fact, a thorough behavioral evaluation in the rodent may be counterproductive, in that it requires so much time that an ability to assess the acute and fluctuating nature of changes is lost. This balance between temporal issues and behavioral thoroughness should be decided ahead of time and aligned with the delirium feature chosen for study.

The second outcome measure would include that reflecting pathophysiology. As discussed earlier and presented in Table 5, this would include various biofluid biomarkers, tissue analysis (eg, enzyme-linked immunosorbent assays, histochemistry, and morphology), electrophysiology, or imaging. These outcomes are critically important as they reflect the primary advantage of animal models: to align the relationship between pathophysiology and behavior in animals, in the hope of translating knowledge to humans. The temporal relationship between behavior and pathophysiology is similarly important, so, to the extent possible, we suggest contemporaneous measurement of behavior and pathophysiology. Secondary outcomes should also leverage knowledge for other disciplines that can help to validate the findings from animal models. For example, microphysiological organ-on-chip technologies can be applied to deconstruct a neuroimmune interaction of relevance to delirium pathogenesis and treatment with human cells.¹²³ Standardized chips and devices can be used with various cell sources,¹²⁴ including human induced pluripotent stem cells (hiPSCs), to define new targets and high-throughput drug discovery applications. Ongoing advances in behavioral assays with the use of artificial intelligence, automated tracking, and machine learning provide novel ways to evaluate complex behaviors, enhancing rigor and reproducibly, and remain largely underutilized in this field.

1.7.7 |. Statistics

Behavioral measures have a degree of subjectivity and, therefore, have a larger coefficient of variation than biofluid biomarkers, for example. Because of this, we suggest that the investigative team incorporate the largest number of animals they can afford or handle, recognizing the costs associated with increasing numbers of animals. Acknowledging that variability and effect size estimates are key metrics in determining the number of animals per group to be studied, we cautiously suggest inclusion of ≥20 animals/group given that effect sizes are rarely known at the time of study inception but are generally small (10–20%). Therefore, behavioral studies with <20 animals per group are likely to be underpowered, whereas this might not be the case with more objective continuous measures like biofluid biomarkers. Ideally, and where possible, a power analysis should be conducted based on the effect size in preliminary data or the literature. When such effect size estimates are unknown, a clinically meaningful effect size could be considered. Analyses should be focused on both individuals and groups rather than only on groups to assess key delirium features like acute onset and fluctuating course.

1.7.8 |. Reporting

Reproducibility and rigor require a priori study design (including a statistical approach), reliable reporting of all details, and depositing of data in publicly available databases after publication (eg, Harvard Dataverse, AlzPed). Access to repositories (including the NIA Biobank and other resources referenced on the NIDUS website < https://deliriumnetwork.org/ >) offer unapparelled opportunities for data mining and for developing bioinformatic resources as encouraged by funding agencies. PREPARE reporting guidelines¹¹⁹ should be followed.

1.8 |. Conclusions and future directions

It is not reasonable to expect any animal model to mimic human delirium with high fidelity. Thus, our overarching recommendation is to utilize animal models to advance our understanding for specific, key features and key pathological hallmarks of delirium as progressively informed by the clinical literature. Animal models may be useful for bidirectional translation of associative studies in humans, followed by more mechanistic studies in animals, and then returning to the human for attempts at mechanistic intervention. As more knowledge about delirium accumulates, animal models should become an integral part of a systems approach to investigating delirium neurobiology, one that interfaces with strategies such as computer neuroscience, machine learning, and artificial intelligence, with the common goal of improving the prevention, diagnosis, and treatment of this complex clinical syndrome. Attention to construct validity (vulnerability and precipitant) as well as face validity (behavior) in the animal is critical to lend credence to the associated pathophysiology. Future work needs to carefully assess and optimize animal physiology during the precipitant phase and to incorporate multiple behavioral paradigms that include a temporal component and analyze the data as individuals rather than groups. Multidisciplinary approaches will likely provide unique opportunities to study delirium in preclinical models. In this regard, NIDUS facilitates the creation of large-scale multidisciplinary projects centered on modeling delirium in animals. The expectation for this and, hopefully, more consortium-based work is that several questions will be tackled in parallel by multidisciplinary teams, overall enhancing rigor and reproducibly for the various protocols and offer designs that better align with clinical randomized controlled trials, as opposed to a single laboratory observation. Delirium, as both an isolated condition and a strong risk factor for AD/ADRD, will continue to impact our society, as has been repeatedly demonstrated by the rapid increase in our aging population and the aftermath of COVID-19. In this regard, it is of paramount importance to advance preclinical research in this field to better understand how predisposing and precipitating factors cause delirium and which interventions will be able to prevent and treat this common neurological complication.

RESEARCH IN CONTEXT.

1. Systematic review: The authors reviewed the literature on animal models for delirium using traditional sources (eg, PubMed). Although human studies of delirium have advanced our understanding of delirium epidemiology and intervention strategies, our knowledge of delirium neurobiology remains limited. Animal models for delirium could advance our knowledge of delirium neurobiology and help develop new prevention and treatment approaches.

2. Integration: Although the application of animal models of delirium remains in a nascent stage, there is a need for rigorous, high-quality science in this area of research to advance understanding of its complex neurobiology.

3. Future directions: This article discusses the application of animal models for delirium, focusing on (a) challenges and strategies for replicating elements of human delirium in animals; (b) utility of fluid, neurophysiology, and neuroimaging markers in animal models; and (c) recommendations for best practices for the development and validation of animal models for delirium.