Prognostic models for prediction of outcomes after traumatic brain injury based on patients admission characteristics.

OBJECTIVE: To identify the best performing prognostic model using admission characteristics to predict mortality at 30 days and functioning outcome at 6-months post-admission in patients with moderate or severe brain injury.

METHODS: Using a retrospective database (n = 1466 patients) of a tertiary trauma care centre, three different models were developed using logistic regression methods for predicting mortality and functioning outcome. The performance of the models was assessed in terms of discrimination and calibration. The models were validated using split sample method. For facilitating clinical usefulness, score charts were derived from the regression models.

RESULTS: The variables motor score, hypotension, pupillary reactivity, age, creatinine level, limb movement (hemiparesis), and tSAH/IVH were found to be the most predictive independent prognostic factors of both mortality and functioning outcome. For both the outcomes, discriminative ability of the three prognostic models was excellent in the development dataset (AUC = 0.845--0.905) as well as the validation data set (AUC = 0.836-0.880). Calibration in the validation data set for model-2 was good (H-L test p-value > 0.05); however, for model-1 and model-3, it was poor (H-L test p-value < 0.05).

CONCLUSION: For clinical decision-making, model-2 is recommended on the basis of good performance in predicting outcomes in patients with moderate or severe TBI in India and other similar countries.