Traumatic cervical spinal cord injury (TCSCI) is a severe and complex trauma that causes profound physical damage and significantly impairs neurological function. The incidence of postoperative delirium (POD) in these patients is alarmingly high, ranging from 18 to 35% (1). Postoperative delirium, an acute cerebral dysfunction syndrome, manifests as altered consciousness, cognitive disturbances, and disrupted sleep–wake cycles (2). In TCSCI patients, POD substantially prolongs hospitalization. While baseline hospital stays may remain stable under normal circumstances, delirium-induced complications—such as behavioral abnormalities, noncompliance with treatment, and the need for intensified nursing care and extended diagnostic/therapeutic interventions—result in significant increases in length of stay (LOS) (3). This not only exacerbates patient suffering but also escalates healthcare costs. Prolonged hospitalization further strains medical resources (e.g., bed occupancy, equipment utilization), challenging healthcare system efficiency and resource allocation (4).

Although widely used delirium prediction tools like the PRE-DELIRIC (PREdiction of DELIRium in ICu patients) model demonstrate moderate predictive value, their development primarily targeted elderly or critically ill populations (5). These models exhibit critical limitations when applied to TCSCI patients due to unaddressed injury-specific factors. TCSCI involves unique spinal cord damage—a pivotal component of the central nervous system—that induces sensory/motor deficits below the injury level and triggers cascading pathophysiological responses (6). The severity of neurological impairment varies markedly across patients, ranging from mild manifestations (e.g., localized hypoesthesia, muscle weakness) to severe cases with complete paraplegia and incontinence. Furthermore, surgical stress responses in TCSCI patients involve distinct mechanisms, where intraoperative trauma, blood loss, and anesthesia collectively exacerbate physiological and psychological stress (7). However, existing models like PRE-DELIRIC fail to incorporate TCSCI-specific predictors such as spinal injury characteristics, neurological deficit severity, or surgical stress parameters, thereby limiting their predictive accuracy in this population (8).

The aforementioned findings highlight the need to establish TCSCI-specific prediction models. Existing delirium prediction tools fail to accurately predict postoperative delirium risk in TCSCI patients due to their inherent limitations (9). A dedicated prediction model would comprehensively integrate TCSCI-specific characteristics, incorporating relevant risk factors (e.g., spinal injury features, severity of neurological deficits, surgical stress responses) and emerging mechanisms such as systemic inflammatory responses, thereby enhancing prediction accuracy and specificity (10). This approach enables early identification of high-risk patients postoperatively, allowing targeted interventions to reduce delirium incidence, shorten hospital stays, alleviate healthcare burdens, and improve patient prognosis (11).

This study aims to construct and validate a TCSCI postoperative delirium risk prediction model by integrating multiple dimensions of patient baseline characteristics, injury characteristics, surgical parameters and biomarkers through a multicenter prospective cohort study, aiming to provide evidence-based basis for early identification of high-risk patients and precise intervention.

Severely traumatized patients frequently experience postoperative delirium (POD) due to perioperative stress responses, inflammatory cascade activation, and neuroendocrine dysregulation, which collectively disrupt central nervous system metabolism (15, 16). POD is a common complication in spinal cord injury patients, significantly prolonging hospitalization, increasing healthcare costs, and correlating with long-term cognitive impairment (17). Despite advancements in perioperative management strategies—such as multimodal analgesia and sleep cycle interventions—the incidence of POD in traumatic cervical spinal cord injury (TCSCI) patients remains notably high (18.6–24.3%) (18–21). Our study observed a POD incidence of 22.08%, It is basically consistent with previous studies.

Our findings can be contextualized within and enrich the broader landscape of POD prediction research. The independent predictors identified in our TCSCI-specific model, such as diabetes mellitus and prolonged operative duration, resonate with findings from large-scale studies across diverse surgical populations. For instance, the individual patient data meta-analysis by Sadeghirad et al. confirmed these as significant risk factors for POD after noncardiac surgery. Conversely, our model revealed distinctive elements; whereas Sadeghirad et al. (22) did not find a significant association for general alcohol consumption, a “history of alcohol abuse” was a strong independent predictor in our cohort, underscoring the importance of delineating specific high-risk behaviors. Furthermore, the incorporation of TCSCI-specific indicators like severe ASIA grades (A-B) provided a tailored precision that may be lacking in models derived from broader surgical populations, even those with high discriminatory power like the machine learning model by Bishara et al. (23). This juxtaposition highlights that while universal risk factors exist, the development of injury-specific models is crucial for accurate prediction in unique, high-risk patient groups like those with TCSCI.

This study identified a significantly elevated incidence of postoperative delirium (POD) in male patients with alcohol abuse (OR = 1.929), potentially attributable to the neurotoxic effects of chronic alcohol consumption. Although the univariate association for alcohol abuse did not meet the extreme threshold of the Bonferroni correction (as detailed in Supplementary material), it emerged as a strong and independent predictor in the multivariate model. This underscores its unique contribution to delirium risk, which is consistent with the established pathophysiology. Chronic alcohol abuse induces frontal lobe atrophy (24) and cortical metabolic suppression (25), impairing executive function and memory integration, thereby increasing delirium susceptibility (26). Furthermore, alcohol-dependent individuals often exhibit adaptive changes in γ-aminobutyric acid (GABA) receptors, necessitating higher intraoperative sedative/analgesic doses. Residual drug effects from delayed metabolism may exacerbate POD risk (27). In elderly TCSCI patients, concurrent spinal stenosis increases the likelihood of occult blood loss from persistent postoperative laminar bone surface oozing. Combined with age-related hematopoietic decline and nutritional deficiencies (due to upper limb mobility impairment), this predisposes patients to chronic anemia (28). Kawaguchi et al. (29) demonstrated that anemia (hemoglobin <10 g/dL or hematocrit <30%) significantly reduces cerebral oxygen delivery, triggering POD through mitochondrial dysfunction and impaired neurotransmitter synthesis. Notably, delirium-associated confusion and dysphagia may worsen nutritional intake, creating a vicious cycle of anemia-delirium-malnutrition. This underscores the clinical imperative to integrate postoperative hemoglobin monitoring and early enteral nutrition into intervention protocols.

A history of diabetes mellitus emerged as an independent POD risk factor (OR = 2.156), with pathophysiology linked to cerebral glucose dysmetabolism and insulin resistance. Studies confirm reduced glucose utilization in the frontal and temporal cortices of diabetic patients (30), leading to synaptic plasticity impairment and cholinergic inhibition, which form the neural basis for executive dysfunction and memory encoding anomalies in delirium (31). In elderly diabetic TCSCI patients, surgical stress combined with glucocorticoid administration may induce acute glycemic fluctuations (>11.1 mmol/L or <3.9 mmol/L). These metabolic derangements activate microglia and proinflammatory cytokine release, amplifying POD risk (32). Therefore, we recommend implementing perioperative glycemic management (target range: 6.0–10.0 mmol/L) and prioritizing analgesic regimens with minimal glycemic impact.

This study confirmed that postoperative delirium (POD) in traumatic cervical spinal cord injury (TCSCI) patients is closely associated with neurological injury severity, perioperative stress intensity, and hemodynamic stability. Patients with ASIA Impairment Scale grades A-B (complete/near-complete injury) exhibited a 3.03-fold higher delirium risk (95% CI: 1.910–4.807) compared to grades C-D patients, likely due to loss of descending sympathetic inhibition in severe spinal injuries, which triggers autonomic dysregulation syndrome. This promotes peripheral inflammatory cytokines (e.g., IL-6, TNF-α) to cross the compromised blood–brain barrier, activating hippocampal microglia and disrupting cholinergic neurotransmission (21, 33). Each additional hour of surgery increased delirium risk by 36.3% (OR = 1.363), reflecting cumulative cerebral metabolic impacts (e.g., prolonged anesthesia suppressing mitochondrial complex IV activity and impairing neuronal ATP synthesis) and intermittent cerebral hypoperfusion caused by reduced vertebral artery flow during cervical hyperextension (34). Notably, intraoperative blood transfusion escalated delirium risk by 147.3% (OR = 2.473), attributed to oxidative stress from free hemoglobin in stored blood and transfusion-related immunomodulation (e.g., CD40L-mediated neuroinflammation) that upregulates MMP-9 expression at the blood–brain barrier, exacerbating central inflammation (35). These findings emphasize the need for comprehensive risk factor integration to refine POD risk stratification and prevention strategies in TCSCI patients. Blood transfusion mediates oxidative stress and neuroinflammation through storage lesion (35), while postoperative hemoglobin (HGB) reduction directly impairs cerebral oxygen delivery (29), independently of transfusion history. Sensitivity analysis further confirmed model robustness (the other factor remained significant when either variable was excluded). This highlights the clinical necessity for coordinated management of transfusion indications and postoperative anemia to achieve comprehensive delirium prevention.

Although previous studies have developed POD prediction models for spinal surgery patients (8, 36–38), this study is the first to establish a specific predictive system for TCSCI patients. The cervical injury model developed by Tamai et al. (8) demonstrated limited predictive efficacy (AUC = 0.66) due to its failure to incorporate perioperative indicators. In contrast, our study significantly improved prediction accuracy (AUC = 0.912) by integrating key variables such as operative duration (OR:1.363) and hemoglobin reduction (OR:0.967). Compared with the nomogram for orthopedic surgery by Fan et al. (38), our model incorporates TCSCI-specific indicators including ASIA grades A-B (OR:3.030) and history of alcohol abuse (OR:1.929), demonstrating unique clinical value. These findings are consistent with the conclusions of Luo et al.’s (37) meta-analysis of 1.1 million spinal surgery cases, which confirmed surgical stress and neurological impairment as core triggers of delirium. Importantly, our TCSCI-specific model shows significantly improved performance over generic models. The current model’s limitations in cognitive assessment mirror those noted in previous studies, suggesting the need for further refinement through multicenter collaboration in future research.

It is noteworthy that while “intraoperative blood transfusion” and “postoperative hemoglobin reduction” are physiologically related (as transfusion is typically administered to correct intraoperative anemia), multivariate regression analysis (Table 2) and collinearity diagnostics (all VIF values <5) demonstrate their independent contributions to POD risk. This suggests two key implications: First, both “intraoperative blood transfusion” and the associated severe surgical trauma are significant risk factors. Second, postoperative hemoglobin levels directly determine early postoperative oxygen delivery capacity, with their reduction constituting a critical pathological basis for delirium induction regardless of transfusion history. Therefore, in clinical practice, alongside evaluating transfusion requirements, close monitoring of postoperative hemoglobin levels and timely correction of anemia are crucial for preventing POD.

This study has several limitations: ① All included cases were sourced from a single medical center, which may introduce selection bias due to regional population characteristics and standardized treatment protocols, potentially limiting the model’s generalizability. Future validation through multicenter prospective cohorts is required to enhance external applicability. ② Although internal validation was performed using Bootstrap resampling, the nomogram’s predictive performance remains untested in external independent datasets. Subsequent plans include multi-regional validation via trauma databases to assess the model’s adaptability to evolving surgical techniques and variations in perioperative management. ③ The age range limitation of this study may affect the universality of the results, which will be further verified by multi-center studies with a wider age spectrum in the future. ④ The potential impact of surgical range (single level vs. multiple levels), Surgical Approach: (Anterior vs. Posterior) was not adjusted in the initial model. Although subsequent analyses showed a significant interaction between surgical range and the primary predictors, this relationship needs to be further validated in future studies with larger sample sizes. ⑤ In the future, external validation is needed through multi-center large sample studies involving more diverse populations to further confirm the generalization ability of the model and optimize its performance. ⑥ This retrospective study did not comprehensively collect or analyze specific details of perioperative anesthesia management (such as types and dosages of anesthetic drugs, depth monitoring data). Although we have argued that variables like operative duration and blood transfusion serve as effective proxies for the overall surgical and anesthetic stress, the absence of direct pharmacological data remains a limitation. Furthermore, while ICU stay was used as an indicator, data on direct postoperative ICU admission was not separately analyzed. These factors are known to influence POD risk. Future prospective studies should systematically document these variables to assess their independent contribution and potential for further enhancing the model’s accuracy. ⑦ Fourth, our variable selection process employed a two-tiered approach. While we performed a Bonferroni correction for univariate analysis (see Supplementary material), the final model inclusion was based on multivariate significance and clinical relevance. This led to the retention of “history of alcohol abuse,” which was a significant independent predictor despite its univariate p-value not surviving the strict correction. We believe this approach enhances the model’s clinical utility, but acknowledge it as a methodological consideration.

The six predictors incorporated into this model (diabetes mellitus, history of alcohol abuse, ASIA grades A-B, operative duration, intraoperative blood transfusion, and postoperative hemoglobin level) all feature easy accessibility and strong objectivity. This grants the model good generalizability and operational practicality even in clinical settings lacking complex anesthetic pharmacologic monitoring. Future prospective studies should systematically collect anesthetic drug data and integrate it with this model to construct a more comprehensive “all-factor” prediction tool.