Animal models of delirium: a narrative review

Abstract

Delirium is an acute cognitive failure characterized by an acute onset and psychomotor disturbances. It is associated with poor outcomes in hospitalized patients, resulting in prolonged hospital stays, elevated rates of mortality, and cognitive dysfunctions after hospital discharge. The pathophysiology of delirium remains poorly understood, making the development of animal models a promising avenue for researchers to elucidate the mechanisms underlying human delirium. These models not only aid in understanding the pathophysiology and etiology of the disease but also enhance the efficacy of nonpharmacological interventions and identify pharmacological therapies. This review aims to present the most relevant and translational animal models of delirium available in the literature and to provide a critical appraisal of these models using contemporary translational benchmarks. In preclinical models, delirium-like behavior is induced in various methods to reproduce the fundamentals of human delirium, including anesthesia, surgery, sepsis, and drug interventions. However, researchers must select a model encompassing face, construct, predictive, and target validity. Delirium currently lacks pharmacological therapies, and this review is to encourage more researchers to delve deeper and shed light on this medical condition, which is highly prevalent in vulnerable and hospitalized patients.

Introduction

Delirium is an acute brain dysfunction characterized by the acute or fluctuating onset of disturbances of consciousness, attention, and cognition. It is associated with increased morbidity, mortality, healthcare costs, and hospital-acquired complications [1, 2]. Thirty percent of older medical patients experience delirium at some time during hospitalization [3, 4]. The delirium rates are likely to be underestimated because many studies of delirium exclude patients with cognitive impairment or dementia at baseline who are particularly vulnerable [5]. Clinically and biologically, delirium coincides with other conditions of acute brain dysfunction, including postoperative neurocognitive disorders (PND) and sepsis-associated encephalopathy (SAE). Outside of clinical trials, delirium is characterized by abrupt onset, fluctuating course, and a short-term departure from baseline mental status. In contrast, PND and SAE are more often marked by cognitive impairments that persist beyond the acute phase of illness or emerge later during recovery. For this review, we included animal studies that modeled early and transient behavioral or neurophysiological changes following an acute insult. Studies using memory-based tasks were viewed as appropriate when assessments were performed within a limited post-insult period and showed a return toward baseline function. In this context, these results are understood as reflecting delirium-like phenomena rather than enduring neurocognitive impairment. Also, we considered the biological mechanisms of delirium, with converging evidence implicating neuroinflammatory processes, alterations in neurotransmission, neuronal stress or injury, and disturbances in cerebral metabolism [6–8]. Mitochondrial dysfunction is also associated with the etiopathogenesis, including mitochondrial DNA (mtDNA) methylation [9]. A dysbalance of the host immune response and neuronal injury are emerging from clinical studies as crucial features of delirium [10, 11], via toll-like receptor 4 (TLR-4) signaling [12], and the activation of many other cellular inflammatory pathways in microglia or the peripheral immune system [7, 9].

Recently, the microbiota-gut-brain axis has also been investigated in relation to delirium pathophysiology, and Parabacteroides distasonis has been associated with POD. Patients who experienced POD exhibited a higher abundance of this bacterium than those with non-delirium [13]. Preclinical studies have demonstrated that administering general anesthesia followed by abdominal surgery with or without infection can significantly modulate the gut microbiome [14–17]. Dysbiosis in this context compromises gut barrier integrity, facilitating the translocation of microbial and non-microbial products into the circulation, heightening systemic cytokine responses, and increasing blood-brain barrier (BBB) permeability [18]. Then, proinflammatory cytokines lead to an imbalance in neurotransmitter levels. Under pathological conditions such as chronic stress, sepsis, aging, or glucocorticoid resistance, hypothalamic-pituitary-adrenal (HPA) axis regulation becomes disrupted. Glucocorticoid receptor (GR) desensitization or downregulation attenuates cortisol’s immunosuppressive efficacy, permitting sustained or elevated expression of proinflammatory cytokines, including interleukin (IL)-1β and IL-6. IL-6, in particular, can further stimulate HPA axis activity, establishing a feed-forward loop that amplifies neuroendocrine and inflammatory signaling [19].

The biological basis of delirium is poorly understood, in part because it is challenging to study severely ill patients with conventional electrophysiologic tests, brain imaging, or neurotransmitter assays. A significant barrier to advancing the understanding of the basic mechanisms of delirium is the lack of adequate animal models. Experimental studies using the delirium model initially helped prove its theoretical basis and replicability. Those models helped establish that acute, reversible disturbances of cognition and alertness could be reproduced [20]. Early models of delirium, predominantly those derived from anticholinergic pharmacologic challenges [20] and EEG [21], proved instrumental in establishing associations between delirium-like states, cholinergic dysfunction, altered consciousness, and EEG slowing. Models using atropine or scopolamine also reproduced behavioral fluctuations, deficits in attention, and changes between hyperactive and hypoactive delirium phenotypes. Early work using EEG as a metric identified correlates of brain function changes during acute cognitive impairment [20–23]. These previous models were fundamental to subsequent models that incorporated systemic inflammation, infection, anesthesia, surgery, aging, and environmental stressors (Fig. 1). The present review builds on this foundational literature by focusing on representative models that more closely align with current translational priorities and the recommendations of the NIDUS Delirium Network. For example, animal studies can directly test the hypotheses, frequently suggested in clinical articles, that peripheral inflammation induces or synergizes with central cholinergic dysfunction to induce transient behavioral change resembling delirium [23]. More sophisticated studies involving combinations of aetiological factors are also possible, and these could model the complexities seen in clinical practice more closely [24]. Animal studies offer the potential to discover novel biomarkers and to evaluate the efficacy of candidate drug treatments [25].

Fig. 1. Conceptual Milestones in the Development of Experimental Delirium Models.

Historical overview of experimental strategies used to model delirium. Early work emphasized cholinergic dysfunction and electrophysiological signatures. Subsequent studies incorporated systemic inflammation and postoperative paradigms, establishing mechanistic links between peripheral immune activation and neurocognitive impairment. Recent advances include attention-specific behavioral paradigms, ICU-like environmental stress models, circadian disruption, and quantitative EEG analyses. The emerging integrative era combines inflammatory and environmental insults with multidimensional phenotyping to improve translational relevance to delirium in critical illness and sepsis.

A pragmatic approach is necessary given the challenge of identifying delirium in experimental animals, including rodents. The goal of animal models for delirium research should be to replicate deficits in brain function resembling symptoms seen in delirious patients (face validity), and these deficits should arise from experimental manipulations that resemble the clinical insults (such as surgery, fracture, sepsis, and infection) that trigger delirium in humans (construct validity) [26]. Achieving face and construct validity would be ideal; however, fulfilling these two criteria appears complex [26].

Concepts such as face, construct, predictive, and target validity are frequently invoked in neuropsychiatric disease modeling, but their application to delirium requires particular caution. Face validity is especially challenging, as key features of delirium, such as subjective awareness, thought disorder, or perceptual disturbance, cannot be directly assessed in animals. Accordingly, commonly used behavioral measures, including reduced open-field exploration (it was used by the original authors to index locomotor activity and anxiety-related behavior, rather than cognition) [27], impaired food-seeking behavior, or altered sleep-wake activity, should be interpreted as indirect proxies for domains relevant to delirium, such as arousal, attention, psychomotor activity, or circadian regulation, rather than as direct analogues of clinical symptoms.

These behavioral readouts are subject to important confounding influences. For example, reduced time spent in the center of an open field may reflect altered arousal or psychomotor slowing, but can also be shaped by anxiety-like behavior or reduced locomotion. Similarly, deficits in food-seeking tasks may index attentional or motivational disruption but can also be influenced by changes in olfaction, appetite, or general health. These limitations highlight the importance of convergent behavioral, electrophysiological, and physiological measures when inferring delirium-like states. A second major difficulty in creating face-valid models of delirium is capturing another prominent feature of delirium: disorganization in space and time. As many behavioral assays rely on voluntary exploration or goal-directed behavior for reward, failure to engage with the task at hand may be attributable to sickness behavior, motivational loss, or anxiety. Some studies have avoided this by designing tasks that animals will work to complete even when conditions are aversive, thereby allowing a clearer dissociation between altered cognition and sickness behavior. Significantly, when investigating these constructs, it is necessary to distinguish between task avoidance and task failure, as both can occur during acute illness. Critically, as delirium is a clinical diagnosis reserved for an acute and fluctuating departure from baseline cognition, face validity will necessarily rely on models of transient cognitive deficits from a pre-insult baseline. Construct validity is generally easier to establish, as many experimental paradigms incorporate clinically relevant triggers such as surgery, infection, systemic inflammation, anesthesia, or stressors that resemble those in the intensive care environment. By contrast, identifying predictive validity remains challenging, in large part because there are currently no approved pharmacological treatments specifically targeting delirium. Some interventions can alleviate delirium-like behaviors in individual models, but these effects have not translated reliably across experimental paradigms or into clinical usefulness. In some experimental settings, anti-inflammatory, metabolic, or cholinergic approaches can alter delirium-like features, but these findings have not consistently translated into clinical benefit, underscoring the ongoing challenges in establishing predictive validity. Target validity remains limited, in part because only a small number of studies have shown that altering a specific molecular or cellular pathway can both trigger and reverse delirium-like states across different experimental models. Although several studies point to specific molecular targets, such as proinflammatory cytokines or cholinergic pathways, relatively few have shown that targeted, reversible manipulation of these pathways can consistently modify delirium-like states across different experimental settings. Taken together, these issues make it clear that no single model captures every aspect of validity. Instead, experimental approaches should be chosen according to the specific biological or mechanistic question being addressed [21, 26, 27]. Together, these themes offer a useful framework for understanding the strengths and limitations of the experimental models discussed below. Animal models of delirium can be performed using anesthetics, surgical procedures, or even drugs. Ideally, animal models that exhibit vulnerability factors, such as age, are equivalent to those observed in humans and are essential for studying delirium [26]. Additionally, it is essential to use assessment tools that capture critical aspects of delirium, as well as multiple behavioral tests and shorter assessment intervals to capture its acute, fluctuating nature. Animal models that reconstruct these etiological processes have the potential to provide a better understanding of the pathophysiology of delirium [26, 28]. This review highlights the most commonly used animal models of delirium, including those based on anesthetic exposure, surgical procedures, sepsis, and cholinergic dysfunction (Table 1). Studies were prioritized when behavioral or neurophysiological assessments were performed during the acute phase (≤7 days) and when models reflected clinically relevant precipitants such as surgery, infection, inflammation, or pharmacological challenge. This context is intended to benchmark included models against delirium phenomenology rather than to provide an exhaustive catalog of all published studies (Table 2). The aim is not to systematically include all studies employing these models, but rather to select representative articles that allow a comprehensive assessment of whether each model fulfills the key behavioral criteria relevant to delirium, as outlined in the NIDUS Delirium Network’s recommendations for future research, acute onset and fluctuating course, inattention, disorganized thinking, altered level of consciousness, memory impairment, psychomotor activity and altered sleep-wake cycle [26] (Fig. 2). Several articles were excluded from this review because their primary focus was on long-term neuropsychiatric outcomes. Since delirium is an acute brain dysfunction, this review arbitrarily defines it as including manuscripts where the behavioral evaluation was performed until seven days after the insult and some other less explored models, such as partial hepatectomy [29], splenectomy [30], urinary catheterization [31], cardiac surgery [32], peripheral surgical wounding [33, 34], EEG head mount surgery [35, 36] were not described here.

Table 1.

Characteristics of animal models of delirium.

Author/Year	Animal	Methodology of delirium induction	Measured behaviour or behaviour surrogate	Main Conclusion
Anesthetics model
Hu et al. [48]	20-month-old male Wistar rats	3.6% sevoflurane for 6 h	Fear conditioning and Y-maze at 1 and 3 days	Impairment of both behaviours at both times
Chen et al. [50]	>18-month-old male and female mice	3.0% sevoflurane for 2 h	Fear conditioning and novel object recognition at 1 day	Impairment of both behaviours
Laparotomy models
Ren et al. [51]	2–8-month-old female C57BL/6J mice	Simple laparotomy under 1.4% isoflurane anaesthesia, followed by 2 h 1.4% isoflurane anaesthesia	New object recognition at 12, 24, and 48 h	Impairment at 24, but not at 12 and 48 h
Peng et al. [52]	4-month-old female C57BL/6J mice	Simple laparotomy under 1.4% isoflurane anaesthesia, followed by 2 h 1.4% isoflurane anaesthesia	Y-maze, open-field, and buried food test at 6, 9, and 24 h	Impairment of the buried food test only at 9 h Impairment of the open field only at 6 h Impairment of the Y-maze at 6 and 9 h
Yang et al. [90]	9 and 18-month-old female C57BL/6J mice	Simple laparotomy under 1.4% isoflurane anaesthesia, followed by 2 h 1.4% isoflurane anaesthesia	Morris Water Maze (training days 4–9 after surgery, test on day 10). Barnes maze (training days 12–15, test on day16)	No impairment in Morris Water Maze Increased latency and escape distance in the Barnes Maze only in older mice
Zhang et al. [55]	2-month-old male C57BL/6J mice	Simple laparotomy under 1.4% isoflurane anaesthesia, followed by 2 h 1.4% isoflurane anaesthesia	Elevated plus maze, open-field, and buried food test at 6, 9, and 24 h	Impairment of all behaviours at 6 and 9 h
Lu et al. [54]	18-month-old female C57BL/6J mice	Simple laparotomy under 1.4% isoflurane anaesthesia, followed by 2 h 1.4% isoflurane anaesthesia	Y-maze, open-field, and buried food test at 6, 9, and 24 h	Impairment of all behaviours at 6 and 9 h
Liufu et al. [53]	18-month-old female C57BL/6J mice	Simple laparotomy under 1.4% isoflurane anaesthesia, followed by 2 h 1.4% isoflurane anaesthesia	Y-maze, open-field, and buried food test at 6, 9, and 24 h	Impairment of all behaviours at 6 and 9 h
Liu et al. [9]	6 and 18-month-old male C57BL/6J mice	Simple laparotomy under 1.4% isoflurane anaesthesia, followed by 2 h 1.4% isoflurane anaesthesia	Y-maze, open field, and buried food and nonselective non-sustained attention tests (NNAT) at 6, 9, and 24 h	Young animals: impairment of the buried food test at 9 h. Impairment of the open field at 6 h. Impairment of the Y-maze at 6 and 9 h. No alteration in NNAT. Aged animals: impairment of the buried food test at all times. Impairment of the open field at all times h. Impairment of the Y-maze at 6 and 9 h. Impairment of NNAT at 24 h
Zhou et al. [56]	18-month-old female C57BL/6J mice	Complex laparotomy under 1.4% isoflurane anaesthesia	Y-maze, open-field, and buried food test at 6, 9, and 24 h	Impairment of the buried food test at 6 h Impairment of the open field at all times Impairment of the open field at 6 and 9 h
Yu et al. [57]	3-month-old male C57BL/6J mice	Complex laparotomy under 3% sevoflurane anaesthesia	Y-maze and open-field at 4 and 24 h	Impairment of the open-field at 4 h Impairment of the Y-maze at 4 and 24 h
Illendula et al. [58]	18–20 month-old male C57BL/6J mice	Complex laparotomy (about 45–60 min) under 3% sevoflurane anaesthesia, followed by an ICU-like environment	Y-maze novel arm preference, buried food, and the attentional set-shifting tests at 0, 6, 12, 18, and 24 h	Impairment of buried food test and attentional set-shifting tests 18 and 24 h Impairment of Y-maze at 0, 6, 18, and 24, but not 18 h
Dulko et al. [59]	18–20 month-old male C57BL/6J mice	Complex laparotomy (about 45–60 min) under 3% sevoflurane anaesthesia, followed by an ICU-like environment	EEG recording and pattern of sleep	EEG slowing, Sleep fragmentation
Orthopedic surgery models
Cibelli et al. [62]	12–14 weeks-old male C57BL6/J mice	Tibial fracture followed by the insertion of a stainless-steel pin into the tibia intramedullary canal, followed by an osteotomy under a 2% isoflurane anaesthesia	Fear conditioning at 3d	Impairment of fear conditioning
Terrando et al. [63]	12–14 weeks-old male C57BL6/J mice	Tibial fracture followed by the insertion of a stainless-steel pin into the tibia intramedullary canal, followed by an osteotomy under a 2% isoflurane anaesthesia	Fear conditioning at 3d	Impairment of fear conditioning
Terrando et al. [64]	12-week-old male C57BL6/J mice	Tibial fracture followed by the insertion of a stainless-steel pin into the tibia intramedullary canal, followed by an osteotomy under a 2% isoflurane anaesthesia	Fear conditioning at 3d LTP 24 and 72 h	Impairment of fear conditioning Impairment of LTP at both times
Xiong et al. [66]	3 and 18-month-old female C57BL6/J mice	Tibial fracture followed by the insertion of a stainless-steel pin into the tibia intramedullary canal, followed by an osteotomy under a 2% isoflurane anaesthesia	Fear conditioning at 3d Long-term potentiation at 24 h	Impairment of fear conditioning and long-term potentiation
Yang et al. [65]	10–14 weeks-old male C57BL6/J mice	Tibial fracture followed by the insertion of a stainless-steel pin into the tibia intramedullary canal, followed by an osteotomy under a 2% isoflurane anaesthesia	Fear conditioning at 3d	Impairment of fear conditioning
Miller-Rhodes et al. [67]	3-month-old male and female C57BL/6J mice	Tibial fracture followed by the insertion of a stainless-steel pin into the tibia intramedullary canal, followed by an osteotomy under a 2% isoflurane anaesthesia	Object recognition test and the memory load object discrimination test at 24 h	Impairment of both tasks
Velagapudi et al. [68]	3-month-old male C57BL/6J mice	Tibial fracture followed by the insertion of a stainless-steel pin into the tibia intramedullary canal, followed by an osteotomy under a 2% isoflurane anaesthesia	5-choice serial reaction time from day 1 to day 7	Impairment from day 1 to day 3
Wang et al. [19]	Three-month-old and 12-month-old CVN-AD mice	Tibial fracture followed by the insertion of a stainless-steel pin into the tibia intramedullary canal, followed by an osteotomy under a 2% isoflurane anaesthesia	5-CSRTT from day 1 to day 7	Impairment at day 1 was higher in CVN-AD (12-month-old) when compared to wild-type mice.
Hua et al. [69]	3-month-old C57BL/6J mice (no sex described)	Tibial fracture followed by the insertion of a stainless-steel pin into the tibia intramedullary canal followed by an osteotomy under a 2% isoflurane anaesthesia	Y-maze and water-maze tasks at 8d	Impairment of both tasks
Sepsis models
Liu et al. [91]	6–8-week-old ICR male mice	Cecal ligation and perforation (once puncture with a 20-gauge needle)	Y-maze and fear conditioning test on days 3, 5, 7	Impairment of both tasks at all times
Xu et al. [79]	8–12-week-old BALB/c male mice	Cecal ligation and perforation (once puncture with a 21-gauge needle)	Novel object recognition test and open field test at day 1 and day 7	Impairment of both tasks at both times Decrease in the time in the center area of the open field
Rashid et al. [76]	6–7-month-old C57BL/6J female mice	Transurethral inoculation of E. coli	Open field and Y-maze at day 3	Impairment of both tasks
Li et al. [78]	8–12-week-old C57BL/6J male mice	Cecal ligation and perforation (twice puncture with a 22-gauge needle)	Open field and Y-maze on days 4 and 7	Impairment of both tasks at both times Decrease in the time in the center area of the open field
Consoli et al. [77]	3-month-old C57BL/6J mice (no sex described)	Cecal slurry (1.5 mg/g) i.p injection	Nest building at 2 and 7d Pre pulse inhibition of acoustic startle response at 4 and 8 h EEG recording from day 1 to 7 LTP at days 2 and 7	Nest building was decreased at day 2. Some animals had a persistent deficit at day 7 Increase in pre-pulse inhibition at both times. Slow-wave EEG at 24 h Decrease LTP at 2d. Animals with persistent nest-building deficits presented also decreased LTP at 7d
Yang et al. [80]	6–8-month-old C57BL/6 male mice	Cecal ligation and perforation (twice puncture with an 18-gauge needle)	Morris water maze and elevated plus maze tasks at day 5	Impairment of both tasks
Anticholinergic drugs
Trzepacz et al. [21]	Wistar rats (no sex or age described – only adult)	Atropine 55 mg/kg bolus i.v. injection	Five-arm maze from 20 to 2900 min EEG recording from 20 to 2900 min	Impaired maze performance until 320 min Reduced EEG frequency and amplitude until 290 min
Leavitt et al. [20]	Wistar rats (no sex or age described – only adult)	Atropine 55, 27.5, 13.5, 6.875, and 3.44 mg/kg bolus i.v. injection	Five-arm maze from 20 to 2900 min EEG recording from 20 to 2900 min	Impaired maze performance until 320 min Reduced EEG frequency and amplitude until 290 min
Tamura et al. [22]	3–6-month-old Male Wistar rats	Biperidene 40 mg/kg single i.p. injection	Simple observation of animal behaviour for 60 min after biperiden EEG recording for 60 min after biperiden	Changes that alternate from hyperactive to hypoactive behavioural EEG desynchronization in the hyperactive state, and EEG slowing in the hypoactive state
Qiu et al. [83]	8-week-old male adult Wistar rats	Scopolamine 1.8 mg/kg single i.p. injection	Open field test and elevated plus maze at 30 min	Abnormal behaviour in all tasks
Qui et al. [84]	2-month-old male SD rats	Scopolamine 0.3, 0.9, 1.2, 1.8 mg/kg single i.p. injection	Open field test, elevated plus maze, and light/dark test time after treatment was not described	Abnormal behaviour only with the higher dose
Cheon et al. [82]	9–12-week-old male C57BL/6 mice	Scopolamine 2 mg/kg single i.p. injection	Open field test, elevated plus maze, and novel object recognition test at 1d	Abnormal behaviour in all tasks

EEG electroencephalogram, ICU intensive care unit, LTP long-term potentiation, i.v. intravenous, i.p. intraperitoneal, OFT open field test, NOR novel object recognition, FC fear conditioning, EPM elevated plus maze, BFT behavioral flexibility test, MWM morris water maze, 5-CSRTT five-choice serial reaction time task, NNAT nonselective non sustained attention test, CVN-AD cholinergic basal forebrain neuron-deficient Alzheimer’s disease.

Table 2.

Behavioral paradigms commonly used in animal models of delirium.

Behavioral	Intended domain (original use)	Brain region	The delirium domain	Strengths	Key limitations/confounds
Barnes maze task	Spatial orientation	Hippocampus, retrosplenial cortex	Spatial disorientation	Strong task engagement despite anxiety	Requires training and repeated trials; less suitable for very early post-insult time windows.
Buried food test	Olfactory-guided motivation	Olfactory bulb, piriform cortex, hypothalamus	Attention/awareness	Sensitive to acute change	Dependent on appetite, olfactory function, and motivation, all of which may be altered during illness or stress.
Elevated plus maze task	Anxiety-related behavior	Amygdala, hippocampus, medial PFC	Psychomotor activity, arousal	Captures anxiety-related behavioral changes	Not specific to delirium; anxiety-related behavior alone does not constitute delirium-like impairment.
Five-choice serial reaction time task (5-CSRTT)	Sustained attention	Medial PFC, anterior cingulate cortex, basal forebrain cholinergic system	Inattention	Direct probe of attentional performance	Food-motivated task; reduced task engagement or non-completion must be distinguished from true attentional deficits.
Nest building test	Goal-directed behavior, executive organization	PFC, hippocampus, hypothalamus	Disorganized thinking, reduced goal-directed behavior	Ethologically relevant; sensitive to acute illness	Strongly influenced by sickness behavior, motivation, and motor capacity; limited specificity for cognitive disorganization.
Novel object recognition test	Recognition memory	Perirhinal cortex, hippocampus	Memory/cognition	Widely used, translational	Often assessed days after insult; may reflect longer-term neurocognitive impairment rather than acute delirium-like states.
Open field test	Locomotor activity, anxiety-related behavior	mPFC, amygdala, striatum, brainstem arousal systems	Psychomotor activity, arousal	Simple, high throughput	Performance is influenced by anxiety, sickness behavior, and reduced locomotion; it does not directly assess cognitive function.
Pre-pulse inhibition test	Sensorimotor gating	PFC, striatum, thalamus, brainstem startle circuits	Attentional filtering, altered awareness	Objective, reflex-based measure with minimal training	Sensitive to stress, arousal state, and hearing; altered gating is not specific to delirium.
Y-maze test	Spatial exploration and working memory	Hippocampus, medial PFC	Attention/cognition	Captures spatial exploration	Influenced by locomotion and motivation, spontaneous alternation reflects exploration rather than attention per se.

Fig. 2. Presence of rodent behavioural tests relevant to delirium as suggested by the recommendations for future research from the NIDUS delirium network in the different models.

Acute onset was defined as a change from baseline in relevant parameters. Inattention was defined as impairment in any of the following tasks: the five-choice serial reaction time task, attentional set-shifting, the buried food test, or nonselective, nonsustained attention tests. Disorganized thinking was defined as impairment in any of the following tasks: nest building, attentional set-shifting, or the T-maze task. Altered level of consciousness was operationalized using behavioral and electrophysiological proxies commonly employed to infer changes in arousal or responsiveness, acknowledging that subjective awareness cannot be directly assessed in animal models. Memory impairment was defined as an impairment in any of these tasks: Spontaneous alternation (Y-maze), T-maze, novel object interest/recognition, nest building, buried food test, and water maze task. Psychomotor activity (agitation or retardation) was defined as an impairment in any of the following tasks: open field to assess time spent in the center or periphery, elevated plus maze, or home cage monitoring. An altered sleep-wake cycle was defined as an impairment in any of these tasks: Home cage monitoring. Created in BioRender (2025).

Anesthetics animal model

Anesthesia is a complex process that utilizes intravenous agents and/or volatile neurochemicals to induce widespread obtundation of neuronal responsiveness and decreased functional interconnectivity [37]. General anesthesia is usually considered a reversible medically induced coma, but there may be subtle, irreversible effects [38]. For a recent review on the effects of anesthetics on brain inflammation, please refer to Dominguini et al. [39]. The impact of even a single exposure to general anesthetics (without surgery) in animals has been shown to induce neural death in areas of cell proliferation, including the hippocampus, in young and older animals [40]. The volatile anesthetic agents sevoflurane and isoflurane have been shown to increase amyloid-beta protein accumulation and induce apoptosis [41]. Sevoflurane is also associated with increased phosphorylated tau (P-tau) in mice and memory deficits. Non-human primates exposed to anesthetics during infancy exhibit abnormalities in emotional reactivity, such as increased anxiety when exposed to strangers [42]. Rhesus monkeys repeatedly exposed to sevoflurane during early infancy develop visual memory deficits that emerge only after the first year of life [43]. Anesthetics alter epigenetic modulation of transcription, demonstrating a global effect on neurodevelopment and synaptogenesis, suggesting that anesthetics can impact the development and function of the nervous system at a molecular level, potentially affecting learning, memory, and behavior [42].

Inhaled anesthetics, such as isoflurane and sevoflurane, are widely used during operative procedures. Prolonged sevoflurane exposure impaired learning and memory, stimulating neuronal apoptosis and synaptic dysfunction in the hippocampus [44, 45]. On the other hand, brief times (between 15 min and two hours) showed a significant cytoprotective effect by attenuating apoptosis induced by ischemia-reperfusion injury [46] or sepsis [47]; therefore, the time and the dose of anesthetics exhibit an essential threshold between damage and neuroprotection. It is worth mentioning that different anesthetics may have different effects on cognitive function and attention levels; therefore, each anesthetic can show specific effects, yet mice of different ages may have different changes in cognitive function following anesthesia [33, 34] and this must be taken into account when using an anesthesia animal model.

We identified only two studies that employed anesthetic agents without surgical procedures to evaluate distinct aspects of delirium relevant to the objectives of our review. A clinically relevant exposure to 3.6% sevoflurane for six hours induced blood–brain barrier (BBB) disruption, likely reflecting endothelial dysfunction and altered tight junction integrity, and was associated with subsequent cognitive impairment at 24 and 72 h post-inhalation [48]. Behavioral phenotyping relied on fear conditioning and the Y-maze test [49], paradigms classically probing hippocampal-dependent associative learning and memory, which were repurposed as proxies for inattention and disorganized thinking. However, the use of these tasks as surrogate measures of delirium-related domains may introduce construct bias, as they do not specifically capture the attentional and arousal disturbances that define delirium. In aged mice, exposure to 3% sevoflurane for 2 h induced neuronal hyperactivity, as measured by in vivo two-photon calcium imaging, that persisted for at least 24 h, and the effect was dependent on endoplasmic reticulum stress pathways. In this study, novel object recognition (a task was used to probe recognition memory via preferential exploration of a novel object, an outcome commonly interpreted as memory performance) and fear conditioning were impaired at 24 h [50]. These findings are summarized in Figs. 2 and 3. In summary, anesthesia-only models are most useful for searching anesthetic-specific effects on brain network function, neuroinflammation, and vulnerability factors, but have central limitations in capturing the multifactorial precipitants and fluctuating clinical course of delirium. In the absence of additional systemic or surgical insults, these paradigms are best viewed as exploratory and mechanistic rather than comprehensive delirium models.

Fig. 3. Characteristics of experimental models used to induce delirium in rodents.

Each model is characterized by the type of agent used, procedure performed, evaluation period, main neuropathological findings such as blood-brain barrier disruption, neuroinflammation, microglial reactivity, and synaptic loss as well as the behavioral tests employed for cognitive assessment. BBB Blood-brain barrier, ICU Intensive Care Unit, CLP Cecal Ligation and Puncture, LPS Lipopolysaccharide, OFT Open Field Test, NOR Novel Object Recognition, FC Fear Conditioning, EPM Elevated Plus Maze, BFT Behavioral Flexibility Test, MWM Morris Water Maze, 5-CSRTT Five-Choice Serial Reaction Time Task. Created in BioRender (2025).

Surgery animal models

Postoperative delirium (POD) is a common complication following surgery, closely associated with both the surgical procedure and anesthesia. A fundamental limitation of POD modeling is the difficulty in isolating the effects of anesthesia and surgery. Nonetheless, such models may still serve as valuable tools for investigating the impact of different anesthetic regimens on POD across various surgical contexts.

Laparotomy models

Probably one of the first studies exploring laparotomy as a construct validity allied to behavioral evaluation (face validity) was published by Ren et al. [51]. The authors performed a simple laparotomy under isoflurane anesthesia and measured new-object exploration time at 12, 24, and 48 h after surgery. An impairment in the performance was observed only at the 12-h time point. In parallel, increased alpha-synuclein and S100 calcium-binding protein B (S100B) in the cortex of those animals was observed; however, a single behavioral test may not adequately assess all aspects of delirium-like behavior.

Over the last decade, a battery of behavioral tests has been used to assess behavioral changes in mice using the laparotomy model Peng et al. [52] described a model of a simple laparotomy under 1.4% isoflurane anesthesia that impaired both the natural and learned behavior of mice with acute onset and fluctuating course in three behavioral tests (Y-maze, open-field, and buried food test (The buried food test assessed olfactory-guided food seeking and motivation, with delayed retrieval more likely reflecting reduced motivation or sensory processing than cognitive impairment) at 6, 9, and 24 h after the procedure). The authors also note a fluctuating pattern of behavior between 6 and 9 h following anesthesia and surgery in animal models, with no such changes observed at the 24-h mark [52]. Although the behavioral battery proposed by Peng et al. [52] marks an important step toward a more multidimensional assessment of postoperative behavioral changes, it also has notable limitations [52]. Using a range of tasks improves sensitivity to acute postoperative changes, but several elements of this battery do not directly investigate cognition. Behaviors interpreted as inattention, disorganized thinking, or cognitive impairment, such as reduced exploration in the open field or poorer performance in food-seeking tasks, can just as plausibly arise from reduced motivation, sickness-related behavior, anxiety, or psychomotor slowing rather than true cognitive dysfunction [26, 28]. Importantly, clearly demonstrating acute and transient cognitive impairment remains a key challenge. Without convincing evidence of short-lived cognitive failure, these behavioral changes may lack the specificity needed to distinguish delirium from other postoperative or illness-related effects [26, 52]. Additionally, this model was further characterized using young and aged (18-month-old) mice, and different behavioural tests such as the water maze, the Barnes maze, and nonselective, non-sustained attention tests [9, 53, 54]. In this context, it was demonstrated that the impairment of delirium-like behavior (open-field test, buried food test, Y-maze, and nonselective, non-sustained attention tests) was accentuated in older animals. These behavioral alterations were accompanied by impairments in hippocampal and prefrontal mitochondrial dynamics, as evidenced by altered levels of disrupted-in-schizophrenia 1 (DISC1), dynamin-related protein 1 (Drp1), and mitofusin 2 (Mfn2), proteins that regulate mitochondrial morphology and function, which in turn were associated with disruption of mitochondrial ultrastructure and epigenetic modulation of mtDNA.

A similar model of laparotomy under isoflurane anesthesia was applied by Zhang [55] to stratify animals as POD susceptible and unsusceptible using standardizing Z scores and hierarchical cluster analysis from three different behavioral tests: open-field, elevated plus maze test, and buried food test early after surgery recovery [55]. In addition to behavioral impairment, hippocampal oxidative stress occurs concurrently with a decrease in ATP levels. Cyclosporine-A improved behavioral alterations and oxidative stress in those animals, suggesting a role for mitochondrial dysfunction in the pathogenesis of POD.

Variations of the simple laparotomy model were proposed by some authors. Zhou et al. [56], during the laparotomy, partially exposed abdominal organs for 2 min before closing the abdominal cavity (this procedure lasted approximately 10 min) in aged mice. Early (6, 9, and 24 h) after the procedure, animals were submitted to the buried food test, open-field, and Y-maze tests, and brain inflammation biomarkers, BBB permeability, and astrocyte and microglia reactivity were evaluated. Surgery induced an impairment on the buried food test at 6, but not 9 and 24 h, an impairment in the OF at all times, and in the Y-maze at 6 and 9 h. Following cognitive impairment, proinflammatory cytokine levels within the brain were observed to elevate, accompanied by BBB dysfunction and glial cell reactivity.

Yu et al. [57] proposed a more complex laparotomy procedure. In this model, animals were submitted to laparotomy of 2.5 cm (under sevoflurane anesthesia) and an exploration of the intestine, which was exteriorized and vigorously rubbed for 30 s. The bowel loops were explored outside the abdominal cavity for 1 min and then returned into the abdominal cavity, mimicking a closer exploratory laparotomy performed in patients. Anxiety, behavioral, and cognitive impairment were observed at 4 and 24 h after surgery. The behavior was associated with hippocampal reactivity of glial cells and increased cytokine content. Additionally, they found a decrease in P-tau at six different sites (S396, S404, T205, S199, AT8, and AT180) and an imbalance in the activity of major related kinases and phosphatases.

The Yu et al, model was further improved using an intensive care unit (ICU)-like environment by Illendula et al. [58]. Older mice (18- to 20-month-old) were submitted to 3 h of sevoflurane anesthesia during which laparotomy (after entering the abdominal cavity, the right and left descending colon, transverse colon, liver, and spleen were gently rubbed with a flexible ear loop for 5 min. The abdominal contents were left exposed to air for an additional 10 min.) was performed. The surgery lasted about 45–60 min. After surgery, animals were sedated for 2 h with propofol (on average, mice received 100 mg/kg i.p. propofol), and then submitted to 12 h of ICU conditions (exposure to 33 W light, 90 dB sounds, and cage-rattling). Afterward, animals were subjected to a battery of behavioral tests (Y-maze novel arm preference, buried food test, attentional set-shifting test, and open-field test) every 6 h for up to 24 h after the end of ICU conditions. A fluctuating performance in the Y-maze was observed between 6 and 24 h. Additionally, from 18 to 24 h, animals exhibited deficits in buried food test performance and attentional set-shifting tests, as well as impairments in attention and cognitive flexibility. A relevant aspect of this study is that the authors categorized animals based on the variation in Z-scores calculated at each time point of behavioral assessment (excluding the open field test). For each animal, the Z-score was computed by subtracting the postoperative test result from the baseline value and dividing the difference by the corresponding standard deviation (SD) of the control group. A Z-score equal to or lower than −1.96 at any time point was considered indicative of delirium-like behavior. To determine the overall incidence of delirium in this model, an animal was classified as exhibiting delirium if it presented at least two Z-scores ≤−1.96 within the 24-h testing period. Another limitation of studies using multidomain behavioral batteries is their dependence on composite Z-scores as a proxy for “delirium severity.” Although Z-score normalization can help combine data across tasks, it may also mask which specific behavioral domains are changing and obscure the factors driving those changes. In particular, tasks such as the open-field test, elevated plus maze, and buried food test are not specific to delirium and are highly influenced by changes in locomotor activity, motivation, sickness behavior, and anxiety. As a result, reduced exploration or food-seeking may reflect diminished motivation or systemic illness rather than acute cognitive impairment. At the same time, performance in anxiety-based tasks such as the elevated plus maze is difficult to interpret without clearly separating anxiety-related effects from attentional or cognitive impairment. Interpreting complex behavioral scores, therefore, requires careful attention to whether deficits observed in individual tasks remain meaningful after accounting for these potential confounds. In this context, experimental paradigms that more directly assess cognitive function, while accounting for motivational or anxiety-related influences, such as spatial navigation tasks, may offer more convincing evidence of true cognitive disorganization. For example, in the Barnes maze, heightened anxiety may increase escape motivation, but impaired performance is more likely to reflect spatial disorientation rather than reduced task engagement, providing a more specific measure of cognitive dysfunction [53]. Taken together, these considerations underscore the importance of transparent reporting and careful, task-level interpretation when composite scores are used to infer delirium-like states. They also show that different proteins associated with synaptic function were decreased in the delirium model (neurotrophic receptor tyrosine kinase1 (NTRK1), syntaxin-1A, α-synuclein, synaptophysin, postsynaptic density protein 95 (PSD-95), and microtubule-associated protein 2 (MAP2). This model was further validated by electroencephalogram (EEG) slowing, sleep fragmentation, and circadian changes [59]. Although highly integrative models capture multiple clinically relevant factors and therefore offer strong construct validity, their complexity can make interpretation challenging. To enhance their value for biological findings, it is important to systematically omit individual components, such as anesthesia, surgical injury, ICU-like stressors, or age, or vary their intensity to clarify their specific contributions. Factorial approaches of this kind can help determine which components most strongly contribute to delirium-like phenotypes and whether observed effects reflect individual insults or their interaction. Such strategies are likely to be critical for interpreting results from complex models into focused mechanistic insights and, ultimately, therapeutic targets. These concepts are summarized in Figs. 2 and 3. In summary, laparotomy-based models offer strong construct validity by mirroring a common clinical trigger of delirium and by capturing acute, time-limited behavioral and physiological changes. At the same time, variability in behavioral outcomes and reliance on indirect measures of cognition underscore persistent challenges in establishing face validity, particularly when trying to distinguish true cognitive dysfunction from sickness behavior or changes in motivation.

Orthopedic surgery models

Numerous human studies have explored the occurrence of delirium after orthopedic surgery. It is estimated that up to 50% of elderly patients suffer delirium after major orthopedic surgery [60, 61]. One of the first studies exploring the tibial fracture model as construct validity allied to behavioral evaluation (face validity) was published by Cibelli et al. [62]. Anesthetized mice were submitted to a tibial fracture, and a stainless-steel pin was inserted into the tibia intramedullary canal, followed by an osteotomy and assessed learning and memory by the fear conditioning test 3 days after surgery. Besides an impairment in the fear conditioning test, operated animals presented an early increase (6 h) of hippocampal proinflammatory cytokines and microglial reactivity. Terrando et al. [63, 64] observed the same results in fear conditioning. These authors also determined an impairment in long-term potentiation (LTP) and brain astrogliosis in operated animals, and the role of TNF-α as an upstream cytokine responsible to brain inflammation. Similar results were observed by Yang et al. [65], which also demonstrated BBB dysfunction and macrophage infiltration in the hippocampus of operated animals. Xiong et al. [66] demonstrated the same impairment in fear conditioning behavior (at three days) and LTP, but using older (18-month-old) mice.

Some authors further explored this model; Miller-Rhodes et al. [67] used two different hippocampal-dependent memory tasks to evaluate brain dysfunction 24 h after surgery. They showed impairment on the “what-where-when” object recognition test (episodic memory) and the memory load object discrimination test. Additionally, a disruption of the BBB and microglial reactivity was demonstrated. Probably, the first study using this model, which evaluated attention at different time points after surgery, was published by Velagapudi [68]. Attention, as evaluated using the five-choice serial reaction time task (5-CSRTT), was significantly impaired over the first three days post-surgery, returning to basal levels from 4 to 7 days after. These alterations were associated with changes in microglial morphology, impaired astrocytic tight junction interactions, and amplified expression of autophagy markers 24 h after surgery. It is significant to note, however, that these attention paradigms are food-motivated and that periods of non-completion of trials, reflecting reduced task engagement, may temporally overlap with failed trials, complicating the interpretation of whether observed deficits reflect inattention per se or diminished motivation during acute illness. Recently, Hua et al. [69] performed different tasks in the tibial fracture model, assessing memory by the Y-maze and water-maze tasks. Both tasks were impaired eight days after surgery, when measured (training started on day two after surgery). Additionally, after 8 days, an increase in proinflammatory cytokines in the hippocampus was observed, followed by decreased brain-derived neurotrophic factor (BDNF) levels and reactive microglia and astrocytes.

Wang et al. [19] further refined this model by employing young and aged APPSwDI/mNos2^−/− Alzheimer’s disease mice (CVN-AD). Older animals exhibited increased brain proinflammatory cytokine levels, BBB dysfunction, and Aβ deposition, accompanied by impaired performance on the 5-CSRTT is used to assess sustained attention and response control by measuring accuracy and omissions during brief visual stimulus detection) attention task. These findings are summarized in Figs. 2 and 3. In summary, orthopedic surgery models offer robust construct validity by engaging inflammatory and stress-related pathways relevant to postoperative delirium and by allowing targeted interrogation of attentional dysfunction. Nonetheless, interpretation of behavioral outcomes is complicated by task engagement and motivational confounds, underscoring the need for careful distinction between task non-completion and true cognitive impairment.

Inflammatory challenge and susceptibility models

Inflammatory challenge paradigms have played a central role in advancing biological models of delirium, particularly by helping differentiate the effects of an acute inflammatory insult from the brain’s pre-existing vulnerabilities. Among these approaches, bacterial lipopolysaccharide (LPS; outer membrane of Gram-negative bacteria) has been extensively used as a consistent innate immune stimulus rather than as a disease model in its own right. LPS helps the investigators study how systemic inflammation interacts with underlying neural susceptibility to produce acute and often fluctuating disturbances in cognition, attention, and arousal that mirror core clinical features of delirium [23]. Cunningham et al. [70] clearly demonstrated that an injection of LPS superimposed on neurodegenerative disease (early stages of prion disease) acutely exacerbates cognitive and motor symptoms of disease that mimic delirium. This research team further established that the interaction between LPS administration and chronic cholinergic hypofunction induced by intracerebroventricular injection of μ-p75-saporin, resulting in depletion of cholinergic neurons in the basal forebrain and reduced hippocampal innervation, leads to acute CNS alterations similar to delirium, thereby providing a biological basis for the cholinergic hypothesis of delirium [23]. In a study using freely moving rats, Mamad and colleagues reported that intraperitoneal administration of LPS (1 mg/kg) induced early, region-specific changes in local field potential activity in the hippocampus and the medial prefrontal cortex. Following LPS administration, hippocampal theta power was evidently reduced, an effect that changes in locomotor activity could not explain. In parallel, delta frequency increased in the hippocampus but not in the prefrontal cortex, indicating a clear regional dissociation. The inflammatory challenge also increased the propensity toward abnormal network synchronization, including the presence of spike-wave discharges that were largely restricted to hippocampal circuits. These data suggest that systemic inflammation can selectively disturb oscillatory dynamics in brain regions essential for attention, memory, and arousal, providing objective electrophysiological markers of inflammation-related brain dysfunction relevant to delirium [71]. O’Neill and colleagues used a susceptibility-focused strategy to test whether prior noradrenergic loss contributes to increased vulnerability to inflammation-related cognitive disturbance. In their study, female mice received a selective lesion of locus coeruleus-derived noradrenaline using the neurotoxin DSP-4, and were challenged two weeks later with a low-dose systemic injection of LPS (100 µg/kg, intraperitoneal) [72]. This intervention produced a substantial reduction in noradrenaline levels in the hippocampus and frontal cortex; it did not exaggerate neuroinflammatory responses, increase sickness behavior, or further impair working memory shortly after the inflammatory challenge. These data propose that noradrenergic depletion alone does not automatically predispose the brain to acute inflammation-induced cognitive dysfunction, highlighting the multifactorial nature of delirium vulnerability and indicating that not all pre-existing neural alterations consistently intensify sensitivity to systemic inflammation [72]. Extending this susceptibility foundation, Kealy and colleagues showed that acute systemic inflammation disrupts brain energy metabolism, contributing to behavioral and cognitive impairment [73]. In mice, systemic administration of LPS at 250 µg/kg induced hypoglycemia, reduced central glucose availability, and blocked spontaneous activity, whereas a lower dose (100 µg/kg) was sufficient to trigger acute cognitive dysfunction in animals with pre-existing neurodegeneration. Further impairment of glucose utilization worsened behavioral and cognitive outcomes, whereas exogenous glucose mitigated these effects, indicating a direct link between bioenergetic stress and brain dysfunction. Also, parallel analyses in individuals revealed changed cerebrospinal fluid glycolytic metabolites in patients experiencing inflammatory trauma-associated delirium. These findings identify disrupted brain energy metabolism as a key mechanism through which systemic inflammation precipitates delirium, particularly in vulnerable brains [73]. In parallel, more reductionist inflammatory approaches have shown that individual cytokines, including IL-1β and TNF-α, can on their own precipitate acute, delirium-like cognitive disturbances when applied in susceptible systems, thereby offering mechanistic insight that complements findings from more complex and clinically emulative models [23, 62, 63]. Additionally, ICU-like models represent a highly translational approach by integrating multiple clinically relevant stressors, including surgery, anesthesia, sleep disruption, and environmental stress. While these models capture several core features of delirium, their complexity necessitates systematic dissection of individual components to clarify the relative contribution of each factor to the overall phenotype.

Sepsis models

Although not universally accepted as a sepsis model, LPS administration remains a valuable tool for delirium modeling. Depending on the dose and route of administration, LPS has been shown to alter various behavioral outcomes, including open field testing, fear conditioning, novel object recognition, and novel location recognition, with effects observed from a few hours up to several days post-administration. LPS-based models have been comprehensively reviewed elsewhere; while discussed here, more in-depth mechanistic insights are available in the existing literature [28, 74, 75].

Sepsis induced by infection, unlike the LPS model, is more widely accepted as a reliable and clinically relevant model. However, a significant limitation of these infection-based models in the context of delirium research is the paucity of studies that have conducted behavioral assessments during the early period following sepsis induction, as most evaluations are performed only after 10 days [75]. Rashid et al. [76] in a model of urinary tract infection, demonstrated an impairment of the performance on the open-field and Y-maze 3 days after infection induction [76]. The cognitive impairment was temporally associated with a hippocampal increase in the content of cleaved caspase-3, which seemed to be driven by IL-6. Recently, a model that more closely mimics sepsis (cecal slurry model) was structured to study delirium [77]. This model did not require surgery or anesthesia (unlike the more usual cecal ligation and puncture (CLP) model), allowing a more acute determination of behavior and EEG recording. Behavior was assessed by the nest-building task and the pre-pulse inhibition of the acoustic startle response (as a behavioral measure of attention and awareness). Cecal slurry significantly decreased nest-building scores, mostly on day 2; however, some animals exhibited a persistent deficit on day 7. Pre-pulse inhibition increased at 4 and 8 h after sepsis induction, suggesting stimulus hypersensitivity. To further characterize the model, it was demonstrated that during the first 24 h after cecal slurry, slow-wave EEG predominance and sleep disruption were observed in those mice.

Using the classical CLP model, it has been demonstrated that sepsis impairs performance in the Y-maze and contextual fear conditioning test from day 3 to day 7 post-induction. CLP mice exhibited BBB disruption and brain edema 24 h after the procedure, along with increased cerebral levels of inflammatory cytokines and oxidative stress markers. Impairments in Y-maze and open-field test performance were also observed on days 3 and 7 following CLP [78]. Additionally, septic animals spent less time in the center of the open field, suggesting reduced psychomotor activity [78]. Cognitive function, as assessed by the novel object recognition test, was also significantly impaired in septic animals at both 1 and 7 days post-CLP [79]. Memory was also assessed using the Morris Water Maze from 1 to 4 days after CLP induction, and task performance was impaired in septic animals. Additionally, the authors evaluated behavior in the Elevated Plus Maze, a classical test of anxiety that can also serve as a proxy for psychomotor activity, an important feature of delirium. The CLP procedure appeared to induce reduced locomotor activity, as evidenced by a decreased number of arm entries [80]. Unfortunately, to the best of our knowledge, other behavioral changes recommended for evaluation in delirium models [26] have not been investigated in the CLP model. These findings are summarized in Figs. 2 and 3.

Anticholinergic drugs

Anticholinergic agents such as atropine and scopolamine have long been suggested to produce delirium-like states in humans and experimental animals [81]. The administration of 55 mg/kg of atropine to Wistar rats induced EEG waves with higher amplitudes and slower frequencies that were paired with cognitive deficits in a five blind alleys maze [21]. A dose-response study of atropine administration explored doses of (55, 27.5, 13.5, 6.875, and 3.44 mg/kg [20]. The lower doses that consistently modified EEG wave amplitudes and frequencies and maze performance were 27.5 mg/kg and 13.5 mg/kg.

Tamura et al. [22] developed a study that evaluated biperiden, a muscarinic receptor antagonist, administration as a delirium model. Biperiden-treated Wistar rats had changes that alternated from hyperactive to hypoactive behavior, like mixed delirium. Rapid walking, excessive random sniffing, and retropulsion were observed in the hyperactive state, alternating with motor arrest and drowsiness in the hypoactive state. These alterations were parallel to EEG desynchronization in the hyperactive state and EEG slowing in the hypoactive state.

Cheon et al. [82] compared surgery and scopolamine models of delirium. Mice in the surgery group underwent abdominal surgery, and scopolamine was administered at a dose of 2 mg/kg. Both models induced a similar and sustained impairment on the elevated plus maze and novel object recognition tasks. An interesting difference between the two models was that surgery did not induce anxiety (measured by the elevated plus maze) nor hyperactive motor activity (measured by the open-field) differently from the scopolamine model. Additionally, scopolamine-treated animals showed neuroinflammatory changes, inflammasome activation, and gene expression patterns relevant to immune/inflammatory reaction and nervous system development, similar to the surgery model. Using similar doses of scopolamine (1.8 mg/kg) Qui et al. [83, 84] demonstrated hyperlocomotion (measured by the open-field) and anxiety behavior in the elevated plus maze and light/dark box. The behavioral changes were accompanied by increased brain and CSF concentrations of monoamine metabolites, activation of NLRP3, and increased IL-1β release. Lower scopolamine doses (0.3, 0.9, and 1.2 mg/kg) did not consistently induce delirium features. It is noteworthy that both scopolamine and atropine induced acute delirium-like behaviors in zebrafish [85]. In this context, zebrafish models should be considered high-throughput, exploratory screening tools, particularly suited for rapid assessment of neurochemical mechanisms and candidate compounds, rather than as standalone models of the delirium syndrome. Due to its characteristics, this model may serve as a valuable tool for screening novel mechanisms and therapeutic agents for delirium [86]. These findings are summarized in Figs. 2 and 3.

Conclusion

The exploration of animal models represents a promising avenue for elucidating the complexities of delirium, offering insights into its underlying mechanisms and potential therapeutic targets. However, to fulfil these objectives, animal models must replicate multiple key features of human delirium, as recommended by the NIDUS Delirium Network [26] (Fig. 2). To date, the laparotomy model most comprehensively fulfils the proposed criteria for an ideal animal model of delirium, followed by the sepsis and orthopaedic surgery models. However, even these models require further replication and validation. For instance, no single study employing the laparotomy model has concurrently assessed all the recommended behavioural and biological parameters. Moreover, it remains unclear whether more complex and prolonged surgical procedures provide additional value compared to simpler approaches. Laparotomy and ICU-environment-based models are emphasised as particularly promising due to their strong construct validity and ability to reproduce multiple clinical precipitants; however, it is important to recognise that evaluation criteria necessarily vary across experimental paradigms. No single animal model currently meets all recommended domains of delirium validity, including face, construct, predictive, and target validity. Rather, each model captures a subset of delirium-relevant features, and their translational value depends on the specific biological or mechanistic question being addressed. Accordingly, laparotomy and ICU-like models are best viewed as complementary tools within a broader experimental framework, rather than as complete representations of the delirium syndrome.

The incorporation of an ICU-like environment into surgical or sepsis models could, in theory, enhance their translational relevance. Nevertheless, available data on this approach are limited. Given that simulating ICU conditions is time-consuming and induces additional stress in animals, further research is needed to determine the true benefit of this addition [87]. It is also conceivable that the ICU environment alone, even in naïve animals, could serve as a standalone model of delirium, though this hypothesis requires empirical support. Among sepsis models, CLP remains the most widely used. However, the fecal slurry model may offer certain advantages and warrants further investigation in the context of delirium research. We strongly encourage the more frequent incorporation of strategies to classify animals as delirium-susceptible or non-susceptible within preclinical models. Although not yet widely adopted, such stratification is likely to improve the reliability and translational relevance of these models. Currently, there is no clear evidence supporting the superiority of one classification method over another, whether through unsupervised cluster analysis, behavioral deviation from baseline or control groups using Z-scores, or other standardization approaches aimed at homogenizing experimental cohorts. Importantly, many of the molecular and cellular alterations observed in delirium models reproduce shared pathways of acute brain stress rather than delirium-specific pathology, underscoring the need to interpret these findings in relation to timing, susceptibility, and reversibility when considering translational relevance. Assumed that delirium is, by definition, a transient deviation from an individual’s baseline cognitive state, study designs that explicitly integrate pre-insult baseline assessment and repeated early post-insult measurements are essential for establishing face validity and for distinguishing delirium-like states from persistent cognitive impairment. This important methodological question should be systematically addressed in future studies. Additionally, we strongly advocate for the more frequent integration of EEG [71, 77, 88] or bispectral EEG [35, 36, 89] recordings alongside behavioural assessments. This approach could significantly enhance the characterization of delirium phenotypes and support the development of targeted therapeutic interventions. Finally, we suggest that simpler models, such as those based solely on anaesthesia, LPS, or anticholinergic administration, should be reserved for addressing highly specific research questions, despite their procedural ease. Consistent with this view, the primary translational value of current delirium models lies in their ability to interrogate mechanisms and test the biological plausibility of interventions, rather than to serve as direct predictors of therapeutic efficacy in humans.

By reproducing critical aspects of human delirium in preclinical settings, researchers can enhance the understanding of its pathophysiology and aetiology, thereby providing a foundation for the development of effective interventions. As the field continues to confront the challenges posed by delirium, sustained efforts are required to advance new insights and therapeutic strategies. Ultimately, this collective endeavour seeks to reduce the burden of delirium on vulnerable hospitalized patients.

Author contributions

TB: Study conception, study design, drafting of report; CSS: Study design, study/research conduct, drafting of report; DD: study/research conduct, drafting of report; CR: Study/research conduct, drafting of report, TN: Drafting of report; GS: Drafting of report; FDP: Study design: study/research conduct: drafting of report. Tatiana Barichello and Carla Simon contributed equally to this manuscript. All of the study authors had full access to all of the data in the study, and can take responsibility for the integrity of the data and the accuracy of the analysis. All authors reviewed the report for important intellectual content and approved the final version.

Funding

Texas Alzheimer’s Research and Care Consortium (TARCC) 2022-26 to TB; The National Football League Players Association (NFLPA) to TB and FDP; NIH/NIA grant R01 AG072491 to TB; and MCTIC/CNPq/FNDCT/MS/SCTIE/DECIT, 07/2020, 401263/2020–7 to FDP.

Competing interests

The authors declare no competing interests.