Development and Internal Validation of a Clinical-Radiological Nomogram for In-Hospital Mortality Prediction in Moderate-to-Severe Traumatic Brain Injury

Abstract

Background:
Traumatic brain injury (TBI) remains a leading cause of in-hospital mortality worldwide, particularly in low- and middle-income countries. Widely used CT-based prognostic systems, such as the Marshall and Rotterdam scores, provide only moderate accuracy and do not integrate key clinical variables. There is a need for a simple, reliable bedside tool that combines clinical and radiological factors for early mortality prediction in moderate-to-severe TBI.

Objective:
To develop and internally validate a clinical–radiological nomogram for predicting in-hospital mortality in adults with moderate-to-severe traumatic brain injury, and to compare its performance with established CT-based scoring systems.

Methods:
This retrospective cohort study included adult patients (≥18 years) with moderate-to-severe TBI (post-resuscitation GCS ≤12) admitted to a tertiary-care trauma center between January 2020 and December 2024. Candidate predictors were selected a priori and included age, admission GCS motor score, pupillary reactivity, Rotterdam CT score, and anticoagulant use. Missing data were handled using multiple imputation. Multivariable logistic regression was used to derive the model, followed by bootstrap internal validation (1,000 resamples) with coefficient shrinkage. Model discrimination, calibration, overall accuracy, and clinical utility were assessed and compared with the Marshall and Rotterdam CT scores.

Results:
A total of 2,400 patients were analyzed; median age was 56 years (IQR 38–73), and 69% were male. In-hospital mortality occurred in 240 patients (10.0%). The final nomogram demonstrated excellent discrimination with an optimism-corrected area under the curve (AUC) of 0.90 (95% CI 0.88–0.92), strong calibration (slope 0.99; intercept 0.00), and a low Brier score (0.062). The nomogram significantly outperformed the Rotterdam CT score (AUC 0.79) and Marshall classification (AUC 0.70) (p<0.001 for both comparisons). Decision curve analysis showed superior net clinical benefit across clinically relevant risk thresholds (10–60%). The model also performed well for predicting unfavorable functional outcome at discharge (AUC 0.86).

Conclusion:
A parsimonious clinical–radiological nomogram incorporating routinely available admission variables accurately predicts in-hospital mortality after moderate-to-severe TBI and outperforms traditional CT-based scoring systems. This tool may aid early risk stratification, prognostication, and clinical decision-making, particularly in resource-limited trauma settings.