Introduction
Traumatic cervical spinal cord injury (CSCI) in older adults has become an increasingly important public health concern in aging societies. In Japan and other developed countries, the incidence of CSCI among individuals aged ≥65 years has increased substantially, reflecting demographic shifts and greater susceptibility to minor trauma such as ground-level falls [1–3]. The treatment strategies for elderly patients with CSCI—whether surgical or conservative—remain controversial. Even in cases with clear surgical indications, neurological outcomes may vary widely [4]. Prognostic assessment is further complicated by comorbidities, reduced physiological reserve, and age-related spinal degeneration [5].
Recent evidence has emphasized the value of magnetic resonance imaging (MRI), particularly fat-suppressed sequences such as short tau inversion recovery (STIR), in evaluating the extent of spinal cord injury. Intramedullary hyperintensity on STIR has been associated with greater tissue damage and poorer neurological outcomes [6,7]. Despite these advances, many studies are limited by small sample sizes or heterogeneous age groups. Few investigations have focused exclusively on elderly patients, whose anatomical and physiological characteristics may influence both imaging findings and outcomes [5,8,9]. Moreover, no large-scale multicenter studies have examined the prognostic significance of STIR signal changes specifically in elderly patients with traumatic CSCI.
The present study therefore focused on patients in whom surgical treatment was considered appropriate and examined the association between initial STIR MRI findings and postoperative neurological outcomes. We hypothesized that a greater extent of intramedullary hyperintensity would predict poorer outcomes and tested this hypothesis using propensity-weighted multivariate analysis.
Materials and Methods
Ethics statement and study design
This multicenter retrospective study was conducted according to the Declaration of Helsinki. The study protocol was approved by the Institutional Review Board (IRB) of Shinshu University School of Medicine (IRB no., 4824) and by the Ethics Committees of all participating institutions. Informed consent was obtained using an opt-out method at each center. The study was designed to evaluate the clinical utility of the extent of STIR hyperintensity as an early and accessible imaging marker for prognosis, rather than to develop a comprehensive prognostic model incorporating all potential predictors.
We collected data from 1,512 patients aged ≥65 years who were hospitalized for traumatic cervical spine or spinal cord injury between 2010 and 2020 across 33 institutions in Japan (Fig. 1). Of these, 602 patients were excluded because they had cervical spine injury without spinal cord involvement or lacked final American Spinal Injury Association (ASIA) classification data. The remaining 910 patients, all of whom underwent surgical treatment, were included in the initial analysis. After excluding an additional 288 cases with missing clinical variables required for propensity score calculation, 622 patients were ultimately included in the final analysis.
Initial MRI examinations were performed at each institution using fat-suppressed STIR sequences. Patients were categorized into three groups according to the extent of intramedullary hyperintensity: H0 (no signal change), H1 (hyperintensity at a single level), and H2 (hyperintensity across multiple levels) (Fig. 2). All participating investigators received written definitions of these categories before data extraction.
Surgical procedures were determined by the treating surgeons at each institution. For analytical purposes, the procedures were classified as posterior decompression, posterior fusion, posterior decompression with fusion, anterior fusion, anterior decompression with fusion, anterior-to-posterior staged surgery, or posterior-to-anterior staged surgery (Supplement 1).
Neurological status was assessed using the American Spinal Injury Association Impairment Scale (AIS) at the time of injury and at the final follow-up (Table 1). A good outcome was defined as achieving AIS grade D or E at final follow-up without deterioration from baseline, or demonstrating an improvement of ≥1 AIS grade. A poor outcome was defined as AIS grade C or worse without improvement from baseline. This definition was chosen to capture clinically meaningful neurological recovery even in severe injuries (e.g., grade A to B), because secondary complications such as pressure ulcers are more common in patients with complete injury [10]. Furthermore, AIS grade conversion (e.g., grade B to C or D) has been used as a clinically relevant endpoint in clinical trials [11].
Baseline characteristics summarized in Table 2 were analyzed as follows: continuous variables were compared using one-way analysis of variance, whereas categorical variables were analyzed using Fisher’s exact test. To adjust for potential confounding and baseline differences among groups, we applied inverse probability of treatment weighting (IPTW) based on generalized propensity score analysis. Variables included in the propensity score model were age, sex, total protein, albumin, hemoglobin, pre-injury activities of daily living, history of comorbidities (e.g., cerebrovascular disease, dementia, diabetes mellitus, hypertension, or heart disease), pre-injury medication use, the presence of associated injuries (e.g., limb, thoracic, or lumbar), trauma characteristics (presence and location of fractures or dislocations), ossified spinal lesions (e.g., ossification of the posterior longitudinal ligament [OPLL], ossification of the ligamentum flavum, or diffuse idiopathic skeletal hyperostosis), and ASIA motor scores.
The C-statistics were 0.830 for H0 versus H1, 0.890 for H0 versus H2, and 0.711 for H1 versus H2. The likelihood of poor outcomes for H0 versus H1, H0 versus H2, and H1 versus H2 was evaluated using weighted logistic regression. IPTW was used to balance baseline covariates between groups, and robust variance estimators were applied to calculate standard errors.
Because data on surgical methods were missing for a substantial proportion of patients in this multicenter registry, surgical procedure was not included in the primary propensity score model. Therefore, we performed complete-case sensitivity analyses (n=393), in which surgical procedure was either added to the generalized propensity score model (Supplement 2) or additionally included as a covariate in the weighted outcome model using a simplified two-category classification (Supplement 3).
A p-value <0.05 was considered statistically significant. Considering the three-group comparison design, p-values were adjusted using the Holm method for the analysis presented in Table 3. All statistical analyses were conducted using R version 4.3.1 (The R Foundation for Statistical Computing, Vienna, Austria).
Results
A total of 910 elderly patients with traumatic CSCI were initially identified. After excluding 288 cases with missing data required for propensity score analysis, 622 patients were included in the final cohort (Fig. 1). The mean age of the cohort was 75.0 years, and 70.7% were male. Based on the presence and extent of intramedullary signal change on STIR images, 132, 350, and 140 patients were classified into the H0, H1, and H2 groups, respectively. The mean follow-up period in the study population was approximately 21 months, with no significant differences among the groups. Surgical method data was available for 393 patients (H0=80, H1=222, H2=91), as summarized in Supplement 1, indicating that posterior procedures were predominant across all groups.
Several baseline characteristics differed among the groups (Table 2). The H0 group demonstrated higher ASIA motor scores and modestly higher total protein and albumin levels, whereas ossified spinal lesions were more frequent in the H2 group (Table 2). Regarding neurological outcomes assessed by the AIS, the rate of good outcomes was 92% in the H0 group, 80% in the H1 group, and 68% in the H2 group (Fig. 3). Analysis of AIS grade distributions at the time of injury and at final follow-up showed that the H0 group had a higher proportion of AIS D at injury and AIS grades D/E at final follow-up, whereas the H2 group exhibited a higher proportion of AIS grades A–C at both time points (Fisher’s exact test, all p<0.001).
Neurological deterioration, defined as a decrease of ≥1 AIS grade from injury to final follow-up, was uncommon in the overall cohort (12/622 [1.9%]) but was observed in all subject groups (H0: 1/132 [0.8%]; H1: 6/350 [1.7%]; H2: 5/140 [3.6%]) (Table 4). Because no patient presented with AIS grade E at the time of injury, deterioration from E to D or worse did not occur; all deterioration events occurred within AIS grades A–D (Table 4).
To evaluate independent associations between MRI findings and neurological outcomes, multivariate logistic regression testing was performed using IPTW to adjust for potential confounders. The analysis demonstrated that both the H1 and H2 groups were significantly associated with poor outcomes compared with the H0 group (Table 3), with weighted odds ratios of 2.4 (95% confidence interval [CI], 1.2–5.0; p=0.040) and 3.2 (95% CI, 1.4–7.2; p=0.014), respectively. Although the crude analysis suggested a trend toward poorer outcomes in the H2 group compared with H1, the weighted difference did not reach significance (p=0.063). Nevertheless, these results indicated a graded relationship between the extent of intramedullary STIR signal change and the risk of a poor neurological outcome.
Discussion
In this large multicenter cohort of elderly patients with traumatic CSCI, we observed a strong positive association between intramedullary STIR hyperintensity and poor neurological outcomes. This relationship remained significant after rigorous multivariate adjustment using IPTW. Our findings are consistent with previous studies by Machino et al. [1] and Miyanji et al. [7], which demonstrated that T2 signal abnormalities correlate with neurological severity and recovery in general spinal cord injury populations. Dobran et al. [12] further emphasized that early MRI findings, particularly patterns of hyperintensity, can help stratify prognosis in patients with cervical trauma.
From a pathological perspective, intramedullary STIR hyperintensity likely reflects a combination of edema, hemorrhage, inflammation, and necrosis [6,13,14]. These secondary injury processes may be exacerbated in elderly individuals because of degenerative spinal canal narrowing, impaired microvascular perfusion, and reduced neuroplasticity [5,8,9]. Notably, Weirich et al. [14] demonstrated that histopathological changes correspond closely with the presence or absence of magnetic resonance signal abnormalities, supporting the role of imaging as a biomarker of injury severity in a rat animal model. Emerging imaging techniques, particularly diffusion tensor imaging (DTI), provide quantitative metrics of white matter integrity and may further refine prognostic models in acute CSCI [15,16]. Integrating DTI with STIR-based grading may therefore improve individualized counselling and surgical planning in older adults.
From a clinical standpoint, the extent of STIR signal change may serve as an early and accessible imaging marker to support initial risk stratification and facilitate discussions with patients and families during early goal setting and rehabilitation planning. This information may also aid surgical decision-making, including considerations regarding the timing and extent of decompression, as suggested by recent studies and guidelines addressing surgical timing in acute CSCI [17,18]. Importantly, neurological recovery after traumatic CSCI is influenced by several established factors, including baseline neurological status and timely surgical decompression within the first 24 hours after injury. Early decompression represents a critical therapeutic window for potential neuroprotection and has been associated with improved neurological outcomes [17,18]. Therefore, STIR-based grading should be interpreted alongside these clinical determinants rather than in isolation.
This study has several limitations. First, although numerous factors are known to influence neurological recovery after traumatic CSCI, the present analysis focused specifically on the prognostic value of STIR signal extent as a pragmatic imaging marker. More comprehensive prognostic modeling that incorporates a broader range of clinical and injury-related variables is being conducted separately within this multicenter registry project; therefore, residual confounding from unmeasured factors cannot be excluded. Second, imaging interpretation was not centralized, and interobserver variability could not be fully assessed. Furthermore, MRI evaluation was based solely on sagittal images, and the timing of image acquisition was not standardized across institutions. Rehabilitation protocols were also not standardized. Third, this study included only surgically treated patients and therefore did not evaluate the comparative effectiveness of surgical versus nonsurgical management. Consequently, the generalizability of these findings is limited to patients for whom surgical treatment is considered appropriate. Finally, although the present study demonstrates strong associations, causal relationships cannot be established and should be confirmed in prospective investigations. Future research should validate these findings through centralized imaging review and explore the integration of MRI biomarkers with functional and electrophysiological data to enable more individualized prognostication. Indeed, Bozzo et al. [19] have emphasized the importance of multimodal approaches that combine imaging, clinical, and laboratory data to improve decision-making in acute CSCI.
Conclusions
This large multicenter study demonstrated that the presence and extent of intramedullary hyperintensity on initial STIR imaging were strongly associated with neurological outcomes in surgically treated elderly patients with traumatic CSCI. These findings support the use of STIR imaging as a practical prognostic marker during the initial evaluation and treatment planning in this high-risk population.
PDF Links
PubReader
XML Download
Full text via DOI
Download Citation
Supplement
