Introduction: Traumatic brain injury (TBI) is a prevalent and debilitating condition. Given its importance, numerous prognostic models exist. However, there is a paucity of models that account for the effects of surgical intervention on mortality. This study aims to develop a prognostic model that accounts for the effects of surgical intervention on inpatient mortality.

Methods: Adult TBI patients from a single institution were retrospectively reviewed to identify predictors of inpatient mortality. Data from patients admitted between 2012 to 2016 was used to develop the model. Subsequently, the model was validated with data from patients admitted between 2017 to 2018. Univariate binary logistic regression was used to identify independent predictors of mortality and multivariate binary logistic regression was used to develop a model to predict inpatient mortality. Receiver operating characteristic (ROC) curve and its area under the curve (AUC) were used to understand the model’s accuracy.

Results: For model development, 1614 TBI patients were included. Among the patients, 57.4% had only TBI, 36.7% had concomitant TBI and other head and neck injuries, 2.5% had concomitant TBI and traumatic spinal injury (TSI) and 2.9% had concomitant TBI, TSI and other head and neck injuries. Other head and neck injuries included fractures, vascular, cranial nerve, laryngeal and pharyngeal injuries. Eleven variables – age, male gender, pulse, systolic blood pressure, Glasgow Coma Scale and Injury Severity Score, skull fracture, absence of ventricles, brainstem involvement, ventilation requirement and surgical intervention were identified by multivariate analysis as significant predictors of outcome. A model was developed using these variables and a ROC was generated. The AUC was 0.936 (95% CI: 0.917, 0.954). When a cut-off value of 0.258 was imposed, its sensitivity was 70.2%, specificity was 94.9%. The Hosmer-Lemeshow Test was insignificant indicating good fit.

To externally validate the model, it was applied on 566 TBI patients admitted between January 2017 to June 2018. Probability of inpatient mortality was calculated using the model and a ROC was generated. The AUC was 0.906 (95% CI: 0.865, 0.946)

Conclusion: A model that quantifies the risk of death with and without surgery will be of great clinical use. Hence, this model has significant potential in helping surgeons make clinical judgements that prioritise patients’ best interests. However, the effect of intervention is best studied in prospective studies. Hence, further studies are needed to validate this model prospectively and in the setting of another institution.

Surgery or no surgery? Developing a prognostic model to ascertain risk of inpatient mortality in patients with traumatic brain injury in Singapore