Comparative Evaluation of Computed Tomography-Based Scoring Systems For Predicting Mortality In Adults With Traumatic Brain Injury: A Retrospective Study

INTRODUCTION

Traumatic brain injury (TBI) is one of the leading causes of morbidity and mortality worldwide, with an annual incidence ranging from 150 to 300 per 100,000 individuals.¹ Severe TBI carries a mortality rate as high as 50%, while moderate TBI accounts for approximately 10% mortality.² Beyond mortality, many survivors experience significant long-term physical, cognitive, and psychosocial disabilities, making TBI a major public health and socioeconomic burden. Accurate early prognostication is therefore critical to guide clinical decision-making, optimise resource allocation, and support informed discussions with patients’ families.³ The aetiology of TBI commonly includes road traffic accidents, falls, and assaults. Prognosis is influenced by several clinical and radiological factors, including hypoxia, age, intracranial pressure (ICP), and systemic complications such as sepsis or pneumonia.⁴ Preventing secondary brain injury, particularly from raised ICP and subsequent herniation, remains a cornerstone of management, as untreated ICP elevation is strongly linked to poor outcomes and mortality.⁵

Among clinical predictors, the Glasgow Coma Scale (GCS) continues to be a widely applied tool, with lower scores correlating with higher mortality. However, neuroimaging-especially computed tomography (CT) has emerged as indispensable.⁶ the acute evaluation of TBI, owing to its rapid acquisition, availability, and ability to detect life-threatening lesions. While magnetic resonance imaging (MRI) offers higher sensitivity for subtle injuries, its limited accessibility and longer acquisition times restrict its use in emergency settings.⁷

Several CT-based scoring systems have been developed to provide a structured, objective assessment of TBI severity. These include the Marshall classification, Rotterdam CT score, Helsinki CT score, Stockholm CT score, and the neuroimaging radiological interpretation system (NIRIS).⁸ They evaluate key imaging features such as midline shift, basal cistern compression, intracranial haemorrhage, and diffuse axonal injury, thereby quantifying injury severity and aiding mortality prediction.⁹ Comparative studies suggest that such structured radiological scores enhance prognostic accuracy beyond clinical variables alone.¹⁰

However, despite their growing adoption internationally, limited research has systematically compared these scoring systems in Indian tertiary care emergency settings. Furthermore, most prior studies have focused on individual scoring systems rather than head-to-head comparisons, leaving uncertainty regarding which systems perform best in specific populations.

MATERIAL AND METHODS

This retrospective observational study was conducted at the tertiary care department. All adult patients (≥18 years) with acute traumatic brain injury (TBI) who presented between 1st January to 30th December 2024 were eligible for inclusion.

RESULTS

1. Clinico-demographical data: The average age was 49.53 years, with 192 males and 86 females. However, mortality is highest in elderly patients, those with severe GCS (50%), raised ICP, and sepsis. Surgery, morbidities and associated injuries were more common among non-survivors [Table 1].

2. Radiological profile: Normal CT findings strongly correlate with survival (p = 0.001). Statistically significant (p < 0.05) findings are associated with higher mortality in diffuse axonal injury, brain haerniation, midline shift, extradural haematoma (EDH), despite being an intracranial bleed, and brain oedema [Table 2 and Graph 1].¹⁴ Diffuse axonal injury and brain oedema are strongly associated with increased mortality (p < 0.001). Skull fracture and EDH showed trends toward significance but did not meet the 0.05 threshold due to patients with EDH being operated on quickly and hence having less mortality.

Close

Graph 1: Above given horizontal bar chart visually compares categorical variables (in this case, CT findings) against a quantitative outcome (mortality rate as a percentage). Each horizontal bar represents a different CT finding, and its length reflects the associated mortality rate.

Export to PPT

3. Scoring systems: Mortality increases with higher scores in all scoring systems expect in the Marshall – II-III has 8.6 % mortality and IV has 4.3%. All scoring systems show statistically significant differences in mortality across their categories (p < 0.001) [Table 3 and Graph 2].¹⁵

Close

Graph 2: Above given forest plot for 278 patients showing the adjusted odds ratios (aORs) for mortality associated with different CT based scoring systems used in traumatic brain injury (TBI), along with their 95% confidence intervals (CIs).

Export to PPT

4. Diagnostic performance metrics of CT scoring systems in predicting mortality: The Marshall score demonstrated the highest sensitivity (93%), but its specificity was comparatively low (32%), resulting in moderate overall accuracy (45%). The Rotterdam score showed sensitivity of 83% and specificity of 51%, achieving the highest accuracy among the systems evaluated (58%). The Stockholm score showed sensitivity of 90%, specificity of 44%, and accuracy of 53%, alongside the largest AUC (0.86). The NIRIS and Helsinki scores yielded intermediate values, both maintaining good sensitivity (>85%) but lower specificity (<42%), corresponding to accuracies of 49% and 47%, respectively [Table 4 and Graph 3].

Close

Graph 3: This chart compares the diagnostic performance metrics of five CT-based traumatic brain injury (TBI) scoring systems. NIRIS – Neuroimaging radiological interpretation system, AUC-Area under the receiver operating characteristic curve.

Export to PPT

5. The receiver-operating characteristic (roc) curve: Analysis for the five CT scoring systems, illustrating their ability to discriminate between survival and mortality in TBI patients [Graph 4].

Close

Graph 4: Each curve plots the true positive rate (Sensitivity) against the false positive rate (1 - Specificity). Stockholm and Rotterdam scores had the highest AUCs, suggesting stronger predictive accuracy. Marshall, NIRIS, and Helsinki followed with moderately strong AUCs.

Export to PPT

6. Mortality: The mortality patterns and causes in a patient population indicate an early mortality rate of 3.58% within the first 3 days due to severe presentation and primary brain injury. The majority of deaths, 35.72%, occur by day 7, with late mortality between 7 days and 1 month primarily caused by complications like infections, diffuse axonal injury [Table 5 and Graph 5].¹⁶ or poor neurological recovery, leading to a total 1-month mortality of 15.1% (42 patients).

Close

Graph 5: This line chart illustrates the mortality trend over the 30-day follow-up period. It highlights that the most critical period for patient mortality is within the first 7 days, where the rate climbs from 2.5% to 11.9%.

Export to PPT

DISCUSSION

In this study of 278 patients with traumatic brain injury (TBI), the overall 30-day mortality was 15.1%, with the majority of deaths occurring within the first week of admission. These findings are consistent with prior literature, which emphasises the critical period of early mortality in TBI patients.¹⁷ Early deaths were largely attributable to primary brain injury and severe clinical presentation, whereas late deaths were more commonly related to secondary complications such as sepsis and raised intracranial pressure.

Limitations

This study has several limitations. Its retrospective design introduces potential biases in data collection and interpretation. Being a single-centre study, generalisability is limited, and the sample size of non-survivors (n=42) restricts subgroup analyses. Biomarkers and advanced imaging modalities, such as MRI, were not included, which might have improved prognostic accuracy.²⁶ Moreover, long-term functional outcomes were not assessed, and mortality alone may not fully capture the burden of TBI.²⁷

Future directions

Future research should focus on prospective multicentre validation of CT-based scoring systems in diverse populations. Integrating radiological scores with clinical variables, pupillary reactivity, and emerging biomarkers could lead to the development of multimodal prognostic models.²⁸ Advances in machine learning may further refine predictive accuracy and allow for individualised patient trajectories.²⁹ Early application of such models in emergency departments has the potential to improve survival and optimise resource utilisation.³⁰

CONCLUSION

This study demonstrates that CT-based scoring systems are reliable tools for predicting mortality in patients with traumatic brain injury. Mortality was strongly associated with advanced age, comorbidities, low Glasgow Coma Scale scores, and complications such as raised intracranial pressure and sepsis. Radiological indicators, including diffuse axonal injury, cerebral oedema, and midline shift, were significant predictors of poor outcomes.

Among the five evaluated CT-based systems, the Stockholm and Helsinki scores exhibited the highest predictive accuracy, supported by their strong area under the curve AUC values and odds ratios. The Rotterdam and Marshall systems also performed well, while NIRIS offered moderate prognostic utility. These findings underscore the value of structured radiological scoring in complementing clinical parameters to enhance early risk stratification.

In resource-constrained emergency settings, incorporation of these scores into routine assessment can aid in timely triage, guide neurosurgical decision-making, and provide families with realistic prognostic information. Moving forward, multicentre prospective studies integrating CT scores with biomarkers and clinical data are warranted to develop multimodal predictive models that could further refine individualised care pathways.